fix: remove downscale from silero vad and use upstream processor

This commit restore the original behavior with frame cutting. While
silero is used on our gpu for files, look like it's not working great on
the live pipeline. To be investigated, but at the moment, what we keep
is:

- refactored to extract the downscale for further processing in the
pipeline
- remove any downscale implementation from audio_chunker and audio_merge
- removed batching from audio_merge too for now

.github/workflows/db_migrations.yml

@@ -17,10 +17,40 @@ on:
 jobs:
   test-migrations:
     runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: postgres:17
+        env:
+          POSTGRES_USER: reflector
+          POSTGRES_PASSWORD: reflector
+          POSTGRES_DB: reflector
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd pg_isready -h 127.0.0.1 -p 5432
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 5
+
+    env:
+      DATABASE_URL: postgresql://reflector:reflector@localhost:5432/reflector
 
     steps:
       - uses: actions/checkout@v4
 
+      - name: Install PostgreSQL client
+        run: sudo apt-get update && sudo apt-get install -y postgresql-client | cat
+
+      - name: Wait for Postgres
+        run: |
+          for i in {1..30}; do
+            if pg_isready -h localhost -p 5432; then
+              echo "Postgres is ready"
+              break
+            fi
+            echo "Waiting for Postgres... ($i)" && sleep 1
+          done
+
       - name: Install uv
         uses: astral-sh/setup-uv@v3
         with:

.github/workflows/deploy.yml

@@ -8,18 +8,30 @@ env:
   ECR_REPOSITORY: reflector
 
 jobs:
-  deploy:
-    runs-on: ubuntu-latest
+  build:
+    strategy:
+      matrix:
+        include:
+          - platform: linux/amd64
+            runner: linux-amd64
+            arch: amd64
+          - platform: linux/arm64
+            runner: linux-arm64
+            arch: arm64
+
+    runs-on: ${{ matrix.runner }}
 
     permissions:
-      deployments: write
       contents: read
 
+    outputs:
+      registry: ${{ steps.login-ecr.outputs.registry }}
+
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
 
       - name: Configure AWS credentials
-        uses: aws-actions/configure-aws-credentials@0e613a0980cbf65ed5b322eb7a1e075d28913a83
+        uses: aws-actions/configure-aws-credentials@v4
         with:
           aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
           aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
@@ -27,21 +39,52 @@ jobs:
 
       - name: Login to Amazon ECR
         id: login-ecr
-        uses: aws-actions/amazon-ecr-login@62f4f872db3836360b72999f4b87f1ff13310f3a
-
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v2
+        uses: aws-actions/amazon-ecr-login@v2
 
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
+        uses: docker/setup-buildx-action@v3
 
-      - name: Build and push
-        id: docker_build
-        uses: docker/build-push-action@v4
+      - name: Build and push ${{ matrix.arch }}
+        uses: docker/build-push-action@v5
         with:
           context: server
-          platforms: linux/amd64,linux/arm64
+          platforms: ${{ matrix.platform }}
           push: true
-          tags: ${{ steps.login-ecr.outputs.registry }}/${{ env.ECR_REPOSITORY }}:latest
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
+          tags: ${{ steps.login-ecr.outputs.registry }}/${{ env.ECR_REPOSITORY }}:latest-${{ matrix.arch }}
+          cache-from: type=gha,scope=${{ matrix.arch }}
+          cache-to: type=gha,mode=max,scope=${{ matrix.arch }}
+          github-token: ${{ secrets.GHA_CACHE_TOKEN }}
+          provenance: false
+
+  create-manifest:
+    runs-on: ubuntu-latest
+    needs: [build]
+
+    permissions:
+      deployments: write
+      contents: read
+
+    steps:
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
+          aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          aws-region: ${{ env.AWS_REGION }}
+
+      - name: Login to Amazon ECR
+        uses: aws-actions/amazon-ecr-login@v2
+
+      - name: Create and push multi-arch manifest
+        run: |
+          # Get the registry URL (since we can't easily access job outputs in matrix)
+          ECR_REGISTRY=$(aws ecr describe-registry --query 'registryId' --output text).dkr.ecr.${{ env.AWS_REGION }}.amazonaws.com
+
+          docker manifest create \
+            $ECR_REGISTRY/${{ env.ECR_REPOSITORY }}:latest \
+            $ECR_REGISTRY/${{ env.ECR_REPOSITORY }}:latest-amd64 \
+            $ECR_REGISTRY/${{ env.ECR_REPOSITORY }}:latest-arm64
+
+          docker manifest push $ECR_REGISTRY/${{ env.ECR_REPOSITORY }}:latest
+
+          echo "✅ Multi-arch manifest pushed: $ECR_REGISTRY/${{ env.ECR_REPOSITORY }}:latest"

.github/workflows/pre-commit.yml

@@ -0,0 +1,24 @@
+name: pre-commit
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  pre-commit:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-python@v5
+      - uses: pnpm/action-setup@v4
+        with:
+          version: 10
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: "pnpm"
+          cache-dependency-path: "www/pnpm-lock.yaml"
+      - name: Install dependencies
+        run: cd www && pnpm install --frozen-lockfile
+      - uses: pre-commit/action@v3.0.1

.github/workflows/test_server.yml

@@ -19,29 +19,41 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - name: Install uv
-        uses: astral-sh/setup-uv@v3
+        uses: astral-sh/setup-uv@v6
         with:
           enable-cache: true
           working-directory: server
-
       - name: Tests
         run: |
           cd server
           uv run -m pytest -v tests
 
-  docker:
-    runs-on: ubuntu-latest
+  docker-amd64:
+    runs-on: linux-amd64
     steps:
       - uses: actions/checkout@v4
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v2
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v2
-      - name: Build and push
-        id: docker_build
-        uses: docker/build-push-action@v4
+        uses: docker/setup-buildx-action@v3
+      - name: Build AMD64
+        uses: docker/build-push-action@v6
         with:
           context: server
-          platforms: linux/amd64,linux/arm64
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
+          platforms: linux/amd64
+          cache-from: type=gha,scope=amd64
+          cache-to: type=gha,mode=max,scope=amd64
+          github-token: ${{ secrets.GHA_CACHE_TOKEN }}
+
+  docker-arm64:
+    runs-on: linux-arm64
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+      - name: Build ARM64
+        uses: docker/build-push-action@v6
+        with:
+          context: server
+          platforms: linux/arm64
+          cache-from: type=gha,scope=arm64
+          cache-to: type=gha,mode=max,scope=arm64
+          github-token: ${{ secrets.GHA_CACHE_TOKEN }}

.gitignore

@@ -14,3 +14,4 @@ data/
 www/REFACTOR.md
 www/reload-frontend
 server/test.sqlite
+CLAUDE.local.md

.pre-commit-config.yaml

@@ -3,10 +3,10 @@
 repos:
   - repo: local
     hooks:
-      - id: yarn-format
-        name: run yarn format
+      - id: format
+        name: run format
         language: system
-        entry: bash -c 'cd www && yarn format'
+        entry: bash -c 'cd www && pnpm format'
         pass_filenames: false
         files: ^www/
 
@@ -23,8 +23,7 @@ repos:
       - id: ruff
         args:
           - --fix
-          - --select
-          - I,F401
+          # Uses select rules from server/pyproject.toml
         files: ^server/
       - id: ruff-format
         files: ^server/

CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## [0.7.2](https://github.com/Monadical-SAS/reflector/compare/v0.7.1...v0.7.2) (2025-08-21)
+
+
+### Bug Fixes
+
+* docker image not loading libgomp.so.1 for torch ([#560](https://github.com/Monadical-SAS/reflector/issues/560)) ([773fccd](https://github.com/Monadical-SAS/reflector/commit/773fccd93e887c3493abc2e4a4864dddce610177))
+* include shared rooms to search ([#558](https://github.com/Monadical-SAS/reflector/issues/558)) ([499eced](https://github.com/Monadical-SAS/reflector/commit/499eced3360b84fb3a90e1c8a3b554290d21adc2))
+
+## [0.7.1](https://github.com/Monadical-SAS/reflector/compare/v0.7.0...v0.7.1) (2025-08-21)
+
+
+### Bug Fixes
+
+* webvtt db null expectation mismatch ([#556](https://github.com/Monadical-SAS/reflector/issues/556)) ([e67ad1a](https://github.com/Monadical-SAS/reflector/commit/e67ad1a4a2054467bfeb1e0258fbac5868aaaf21))
+
+## [0.7.0](https://github.com/Monadical-SAS/reflector/compare/v0.6.1...v0.7.0) (2025-08-21)
+
+
+### Features
+
+* delete recording with transcript ([#547](https://github.com/Monadical-SAS/reflector/issues/547)) ([99cc984](https://github.com/Monadical-SAS/reflector/commit/99cc9840b3f5de01e0adfbfae93234042d706d13))
+* pipeline improvement with file processing, parakeet, silero-vad ([#540](https://github.com/Monadical-SAS/reflector/issues/540)) ([bcc29c9](https://github.com/Monadical-SAS/reflector/commit/bcc29c9e0050ae215f89d460e9d645aaf6a5e486))
+* postgresql migration and removal of sqlite in pytest ([#546](https://github.com/Monadical-SAS/reflector/issues/546)) ([cd1990f](https://github.com/Monadical-SAS/reflector/commit/cd1990f8f0fe1503ef5069512f33777a73a93d7f))
+* search backend ([#537](https://github.com/Monadical-SAS/reflector/issues/537)) ([5f9b892](https://github.com/Monadical-SAS/reflector/commit/5f9b89260c9ef7f3c921319719467df22830453f))
+* search frontend  ([#551](https://github.com/Monadical-SAS/reflector/issues/551)) ([3657242](https://github.com/Monadical-SAS/reflector/commit/365724271ca6e615e3425125a69ae2b46ce39285))
+
+
+### Bug Fixes
+
+* evaluation cli event wrap ([#536](https://github.com/Monadical-SAS/reflector/issues/536)) ([941c3db](https://github.com/Monadical-SAS/reflector/commit/941c3db0bdacc7b61fea412f3746cc5a7cb67836))
+* use structlog not logging ([#550](https://github.com/Monadical-SAS/reflector/issues/550)) ([27e2f81](https://github.com/Monadical-SAS/reflector/commit/27e2f81fda5232e53edc729d3e99c5ef03adbfe9))
+
+## [0.6.1](https://github.com/Monadical-SAS/reflector/compare/v0.6.0...v0.6.1) (2025-08-06)
+
+
+### Bug Fixes
+
+* delayed waveform loading ([#538](https://github.com/Monadical-SAS/reflector/issues/538)) ([ef64146](https://github.com/Monadical-SAS/reflector/commit/ef64146325d03f64dd9a1fe40234fb3e7e957ae2))
+
 ## [0.6.0](https://github.com/Monadical-SAS/reflector/compare/v0.5.0...v0.6.0) (2025-08-05)
 
 

CLAUDE.md

@@ -62,7 +62,7 @@ uv run python -m reflector.tools.process path/to/audio.wav
 **Setup:**
 ```bash
 # Install dependencies
-yarn install
+pnpm install
 
 # Copy configuration templates
 cp .env_template .env
@@ -72,19 +72,19 @@ cp config-template.ts config.ts
 **Development:**
 ```bash
 # Start development server
-yarn dev
+pnpm dev
 
 # Generate TypeScript API client from OpenAPI spec
-yarn openapi
+pnpm openapi
 
 # Lint code
-yarn lint
+pnpm lint
 
 # Format code
-yarn format
+pnpm format
 
 # Build for production
-yarn build
+pnpm build
 ```
 
 ### Docker Compose (Full Stack)

README.md

@@ -79,7 +79,7 @@ Start with `cd www`.
 **Installation**
 
 ```bash
-yarn install
+pnpm install
 cp .env_template .env
 cp config-template.ts config.ts
 ```
@@ -89,7 +89,7 @@ Then, fill in the environment variables in `.env` and the configuration in `conf
 **Run in development mode**
 
 ```bash
-yarn dev
+pnpm dev
 ```
 
 Then (after completing server setup and starting it) open [http://localhost:3000](http://localhost:3000) to view it in the browser.
@@ -99,7 +99,7 @@ Then (after completing server setup and starting it) open [http://localhost:3000
 To generate the TypeScript files from the openapi.json file, make sure the python server is running, then run:
 
 ```bash
-yarn openapi
+pnpm openapi
 ```
 
 ### Backend

compose.yml

@@ -39,11 +39,12 @@ services:
     image: node:18
     ports:
       - "3000:3000"
-    command: sh -c "yarn install && yarn dev"
+    command: sh -c "corepack enable && pnpm install && pnpm dev"
     restart: unless-stopped
     working_dir: /app
     volumes:
       - ./www:/app/
+      - /app/node_modules
     env_file:
       - ./www/.env.local
 

server/.gitignore

@@ -176,7 +176,8 @@ artefacts/
 audio_*.wav
 
 # ignore local database
-reflector.sqlite3
+*.sqlite3
+*.db
 data/
 
 dump.rdb

server/Dockerfile

@@ -1,7 +1,8 @@
 FROM python:3.12-slim
 
 ENV PYTHONUNBUFFERED=1 \
-    UV_LINK_MODE=copy
+    UV_LINK_MODE=copy \
+    UV_NO_CACHE=1
 
 # builder install base dependencies
 WORKDIR /tmp
@@ -13,8 +14,8 @@ ENV PATH="/root/.local/bin/:$PATH"
 # install application dependencies
 RUN mkdir -p /app
 WORKDIR /app
-COPY pyproject.toml uv.lock /app/
-RUN touch README.md && env uv sync --compile-bytecode --locked
+COPY pyproject.toml uv.lock README.md /app/
+RUN uv sync --compile-bytecode --locked
 
 # pre-download nltk packages
 RUN uv run python -c "import nltk; nltk.download('punkt_tab'); nltk.download('averaged_perceptron_tagger_eng')"
@@ -26,4 +27,15 @@ COPY migrations /app/migrations
 COPY reflector /app/reflector
 WORKDIR /app
 
+# Create symlink for libgomp if it doesn't exist (for ARM64 compatibility)
+RUN if [ "$(uname -m)" = "aarch64" ] && [ ! -f /usr/lib/libgomp.so.1 ]; then \
+    LIBGOMP_PATH=$(find /app/.venv/lib -path "*/torch.libs/libgomp*.so.*" 2>/dev/null | head -n1); \
+    if [ -n "$LIBGOMP_PATH" ]; then \
+    ln -sf "$LIBGOMP_PATH" /usr/lib/libgomp.so.1; \
+    fi \
+    fi
+
+# Pre-check just to make sure the image will not fail
+RUN uv run python -c "import silero_vad.model"
+
 CMD ["./runserver.sh"]

server/README.md

@@ -40,3 +40,5 @@ uv run python -c "from reflector.pipelines.main_live_pipeline import task_pipeli
 ```bash
 uv run python -c "from reflector.pipelines.main_live_pipeline import pipeline_post; pipeline_post(transcript_id='TRANSCRIPT_ID')"
 ```
+
+.

server/gpu/modal_deployments/README.md

@@ -4,7 +4,8 @@ This repository hold an API for the GPU implementation of the Reflector API serv
 and use [Modal.com](https://modal.com)
 
 - `reflector_diarizer.py` - Diarization API
-- `reflector_transcriber.py` - Transcription API
+- `reflector_transcriber.py` - Transcription API (Whisper)
+- `reflector_transcriber_parakeet.py` - Transcription API (NVIDIA Parakeet)
 - `reflector_translator.py` - Translation API
 
 ## Modal.com deployment
@@ -19,6 +20,10 @@ $ modal deploy reflector_transcriber.py
 ...
 └── 🔨 Created web => https://xxxx--reflector-transcriber-web.modal.run
 
+$ modal deploy reflector_transcriber_parakeet.py
+...
+└── 🔨 Created web => https://xxxx--reflector-transcriber-parakeet-web.modal.run
+
 $ modal deploy reflector_llm.py
 ...
 └── 🔨 Created web => https://xxxx--reflector-llm-web.modal.run
@@ -68,6 +73,86 @@ Authorization: bearer <REFLECTOR_APIKEY>
 
 ### Transcription
 
+#### Parakeet Transcriber (`reflector_transcriber_parakeet.py`)
+
+NVIDIA Parakeet is a state-of-the-art ASR model optimized for real-time transcription with superior word-level timestamps.
+
+**GPU Configuration:**
+- **A10G GPU** - Used for `/v1/audio/transcriptions` endpoint (small files, live transcription)
+  - Higher concurrency (max_inputs=10)
+  - Optimized for multiple small audio files
+  - Supports batch processing for efficiency
+
+- **L40S GPU** - Used for `/v1/audio/transcriptions-from-url` endpoint (large files)
+  - Lower concurrency but more powerful processing
+  - Optimized for single large audio files
+  - VAD-based chunking for long-form audio
+
+##### `/v1/audio/transcriptions` - Small file transcription
+
+**request** (multipart/form-data)
+- `file` or `files[]` - audio file(s) to transcribe
+- `model` - model name (default: `nvidia/parakeet-tdt-0.6b-v2`)
+- `language` - language code (default: `en`)
+- `batch` - whether to use batch processing for multiple files (default: `true`)
+
+**response**
+```json
+{
+    "text": "transcribed text",
+    "words": [
+        {"word": "hello", "start": 0.0, "end": 0.5},
+        {"word": "world", "start": 0.5, "end": 1.0}
+    ],
+    "filename": "audio.mp3"
+}
+```
+
+For multiple files with batch=true:
+```json
+{
+    "results": [
+        {
+            "filename": "audio1.mp3",
+            "text": "transcribed text",
+            "words": [...]
+        },
+        {
+            "filename": "audio2.mp3",
+            "text": "transcribed text",
+            "words": [...]
+        }
+    ]
+}
+```
+
+##### `/v1/audio/transcriptions-from-url` - Large file transcription
+
+**request** (application/json)
+```json
+{
+    "audio_file_url": "https://example.com/audio.mp3",
+    "model": "nvidia/parakeet-tdt-0.6b-v2",
+    "language": "en",
+    "timestamp_offset": 0.0
+}
+```
+
+**response**
+```json
+{
+    "text": "transcribed text from large file",
+    "words": [
+        {"word": "hello", "start": 0.0, "end": 0.5},
+        {"word": "world", "start": 0.5, "end": 1.0}
+    ]
+}
+```
+
+**Supported file types:** mp3, mp4, mpeg, mpga, m4a, wav, webm
+
+#### Whisper Transcriber (`reflector_transcriber.py`)
+
 `POST /transcribe`
 
 **request** (multipart/form-data)

server/gpu/modal_deployments/reflector_diarizer.py

@@ -4,14 +4,80 @@ Reflector GPU backend - diarizer
 """
 
 import os
+import uuid
+from typing import Mapping, NewType
+from urllib.parse import urlparse
 
-import modal.gpu
-from modal import App, Image, Secret, asgi_app, enter, method
-from pydantic import BaseModel
+import modal
 
 PYANNOTE_MODEL_NAME: str = "pyannote/speaker-diarization-3.1"
 MODEL_DIR = "/root/diarization_models"
-app = App(name="reflector-diarizer")
+UPLOADS_PATH = "/uploads"
+SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
+
+DiarizerUniqFilename = NewType("DiarizerUniqFilename", str)
+AudioFileExtension = NewType("AudioFileExtension", str)
+
+app = modal.App(name="reflector-diarizer")
+
+# Volume for temporary file uploads
+upload_volume = modal.Volume.from_name("diarizer-uploads", create_if_missing=True)
+
+
+def detect_audio_format(url: str, headers: Mapping[str, str]) -> AudioFileExtension:
+    parsed_url = urlparse(url)
+    url_path = parsed_url.path
+
+    for ext in SUPPORTED_FILE_EXTENSIONS:
+        if url_path.lower().endswith(f".{ext}"):
+            return AudioFileExtension(ext)
+
+    content_type = headers.get("content-type", "").lower()
+    if "audio/mpeg" in content_type or "audio/mp3" in content_type:
+        return AudioFileExtension("mp3")
+    if "audio/wav" in content_type:
+        return AudioFileExtension("wav")
+    if "audio/mp4" in content_type:
+        return AudioFileExtension("mp4")
+
+    raise ValueError(
+        f"Unsupported audio format for URL: {url}. "
+        f"Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
+    )
+
+
+def download_audio_to_volume(
+    audio_file_url: str,
+) -> tuple[DiarizerUniqFilename, AudioFileExtension]:
+    import requests
+    from fastapi import HTTPException
+
+    print(f"Checking audio file at: {audio_file_url}")
+    response = requests.head(audio_file_url, allow_redirects=True)
+    if response.status_code == 404:
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    print(f"Downloading audio file from: {audio_file_url}")
+    response = requests.get(audio_file_url, allow_redirects=True)
+
+    if response.status_code != 200:
+        print(f"Download failed with status {response.status_code}: {response.text}")
+        raise HTTPException(
+            status_code=response.status_code,
+            detail=f"Failed to download audio file: {response.status_code}",
+        )
+
+    audio_suffix = detect_audio_format(audio_file_url, response.headers)
+    unique_filename = DiarizerUniqFilename(f"{uuid.uuid4()}.{audio_suffix}")
+    file_path = f"{UPLOADS_PATH}/{unique_filename}"
+
+    print(f"Writing file to: {file_path} (size: {len(response.content)} bytes)")
+    with open(file_path, "wb") as f:
+        f.write(response.content)
+
+    upload_volume.commit()
+    print(f"File saved as: {unique_filename}")
+    return unique_filename, audio_suffix
 
 
 def migrate_cache_llm():
@@ -39,7 +105,7 @@ def download_pyannote_audio():
 
 
 diarizer_image = (
-    Image.debian_slim(python_version="3.10.8")
+    modal.Image.debian_slim(python_version="3.10.8")
     .pip_install(
         "pyannote.audio==3.1.0",
         "requests",
@@ -55,7 +121,8 @@ diarizer_image = (
         "hf-transfer",
     )
     .run_function(
-        download_pyannote_audio, secrets=[Secret.from_name("my-huggingface-secret")]
+        download_pyannote_audio,
+        secrets=[modal.Secret.from_name("hf_token")],
     )
     .run_function(migrate_cache_llm)
     .env(
@@ -70,44 +137,51 @@ diarizer_image = (
 
 
 @app.cls(
-    gpu=modal.gpu.A100(size="40GB"),
+    gpu="A100",
     timeout=60 * 30,
-    scaledown_window=60,
-    allow_concurrent_inputs=1,
     image=diarizer_image,
+    volumes={UPLOADS_PATH: upload_volume},
+    enable_memory_snapshot=True,
+    experimental_options={"enable_gpu_snapshot": True},
+    secrets=[
+        modal.Secret.from_name("hf_token"),
+    ],
 )
+@modal.concurrent(max_inputs=1)
 class Diarizer:
-    @enter()
+    @modal.enter(snap=True)
     def enter(self):
         import torch
         from pyannote.audio import Pipeline
 
         self.use_gpu = torch.cuda.is_available()
         self.device = "cuda" if self.use_gpu else "cpu"
+        print(f"Using device: {self.device}")
         self.diarization_pipeline = Pipeline.from_pretrained(
-            PYANNOTE_MODEL_NAME, cache_dir=MODEL_DIR
+            PYANNOTE_MODEL_NAME,
+            cache_dir=MODEL_DIR,
+            use_auth_token=os.environ["HF_TOKEN"],
         )
         self.diarization_pipeline.to(torch.device(self.device))
 
-    @method()
-    def diarize(self, audio_data: str, audio_suffix: str, timestamp: float):
-        import tempfile
-
+    @modal.method()
+    def diarize(self, filename: str, timestamp: float = 0.0):
         import torchaudio
 
-        with tempfile.NamedTemporaryFile("wb+", suffix=f".{audio_suffix}") as fp:
-            fp.write(audio_data)
+        upload_volume.reload()
 
-            print("Diarizing audio")
-            waveform, sample_rate = torchaudio.load(fp.name)
+        file_path = f"{UPLOADS_PATH}/{filename}"
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(f"File not found: {file_path}")
+
+        print(f"Diarizing audio from: {file_path}")
+        waveform, sample_rate = torchaudio.load(file_path)
         diarization = self.diarization_pipeline(
             {"waveform": waveform, "sample_rate": sample_rate}
         )
 
         words = []
-            for diarization_segment, _, speaker in diarization.itertracks(
-                yield_label=True
-            ):
+        for diarization_segment, _, speaker in diarization.itertracks(yield_label=True):
             words.append(
                 {
                     "start": round(timestamp + diarization_segment.start, 3),
@@ -127,17 +201,18 @@ class Diarizer:
 @app.function(
     timeout=60 * 10,
     scaledown_window=60 * 3,
-    allow_concurrent_inputs=40,
     secrets=[
-        Secret.from_name("reflector-gpu"),
+        modal.Secret.from_name("reflector-gpu"),
     ],
+    volumes={UPLOADS_PATH: upload_volume},
     image=diarizer_image,
 )
-@asgi_app()
+@modal.concurrent(max_inputs=40)
+@modal.asgi_app()
 def web():
-    import requests
     from fastapi import Depends, FastAPI, HTTPException, status
     from fastapi.security import OAuth2PasswordBearer
+    from pydantic import BaseModel
 
     diarizerstub = Diarizer()
 
@@ -153,35 +228,26 @@ def web():
                 headers={"WWW-Authenticate": "Bearer"},
             )
 
-    def validate_audio_file(audio_file_url: str):
-        # Check if the audio file exists
-        response = requests.head(audio_file_url, allow_redirects=True)
-        if response.status_code == 404:
-            raise HTTPException(
-                status_code=response.status_code,
-                detail="The audio file does not exist.",
-            )
-
     class DiarizationResponse(BaseModel):
         result: dict
 
-    @app.post(
-        "/diarize", dependencies=[Depends(apikey_auth), Depends(validate_audio_file)]
-    )
-    def diarize(
-        audio_file_url: str, timestamp: float = 0.0
-    ) -> HTTPException | DiarizationResponse:
-        # Currently the uploaded files are in mp3 format
-        audio_suffix = "mp3"
-
-        print("Downloading audio file")
-        response = requests.get(audio_file_url, allow_redirects=True)
-        print("Audio file downloaded successfully")
+    @app.post("/diarize", dependencies=[Depends(apikey_auth)])
+    def diarize(audio_file_url: str, timestamp: float = 0.0) -> DiarizationResponse:
+        unique_filename, audio_suffix = download_audio_to_volume(audio_file_url)
 
+        try:
             func = diarizerstub.diarize.spawn(
-            audio_data=response.content, audio_suffix=audio_suffix, timestamp=timestamp
+                filename=unique_filename, timestamp=timestamp
             )
             result = func.get()
             return result
+        finally:
+            try:
+                file_path = f"{UPLOADS_PATH}/{unique_filename}"
+                print(f"Deleting file: {file_path}")
+                os.remove(file_path)
+                upload_volume.commit()
+            except Exception as e:
+                print(f"Error cleaning up {unique_filename}: {e}")
 
     return app

server/gpu/modal_deployments/reflector_transcriber_parakeet.py

@@ -0,0 +1,622 @@
+import logging
+import os
+import sys
+import threading
+import uuid
+from typing import Mapping, NewType
+from urllib.parse import urlparse
+
+import modal
+
+MODEL_NAME = "nvidia/parakeet-tdt-0.6b-v2"
+SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
+SAMPLERATE = 16000
+UPLOADS_PATH = "/uploads"
+CACHE_PATH = "/cache"
+VAD_CONFIG = {
+    "max_segment_duration": 30.0,
+    "batch_max_files": 10,
+    "batch_max_duration": 5.0,
+    "min_segment_duration": 0.02,
+    "silence_padding": 0.5,
+    "window_size": 512,
+}
+
+ParakeetUniqFilename = NewType("ParakeetUniqFilename", str)
+AudioFileExtension = NewType("AudioFileExtension", str)
+
+app = modal.App("reflector-transcriber-parakeet")
+
+# Volume for caching model weights
+model_cache = modal.Volume.from_name("parakeet-model-cache", create_if_missing=True)
+# Volume for temporary file uploads
+upload_volume = modal.Volume.from_name("parakeet-uploads", create_if_missing=True)
+
+image = (
+    modal.Image.from_registry(
+        "nvidia/cuda:12.8.0-cudnn-devel-ubuntu22.04", add_python="3.12"
+    )
+    .env(
+        {
+            "HF_HUB_ENABLE_HF_TRANSFER": "1",
+            "HF_HOME": "/cache",
+            "DEBIAN_FRONTEND": "noninteractive",
+            "CXX": "g++",
+            "CC": "g++",
+        }
+    )
+    .apt_install("ffmpeg")
+    .pip_install(
+        "hf_transfer==0.1.9",
+        "huggingface_hub[hf-xet]==0.31.2",
+        "nemo_toolkit[asr]==2.3.0",
+        "cuda-python==12.8.0",
+        "fastapi==0.115.12",
+        "numpy<2",
+        "librosa==0.10.1",
+        "requests",
+        "silero-vad==5.1.0",
+        "torch",
+    )
+    .entrypoint([])  # silence chatty logs by container on start
+)
+
+
+def detect_audio_format(url: str, headers: Mapping[str, str]) -> AudioFileExtension:
+    parsed_url = urlparse(url)
+    url_path = parsed_url.path
+
+    for ext in SUPPORTED_FILE_EXTENSIONS:
+        if url_path.lower().endswith(f".{ext}"):
+            return AudioFileExtension(ext)
+
+    content_type = headers.get("content-type", "").lower()
+    if "audio/mpeg" in content_type or "audio/mp3" in content_type:
+        return AudioFileExtension("mp3")
+    if "audio/wav" in content_type:
+        return AudioFileExtension("wav")
+    if "audio/mp4" in content_type:
+        return AudioFileExtension("mp4")
+
+    raise ValueError(
+        f"Unsupported audio format for URL: {url}. "
+        f"Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
+    )
+
+
+def download_audio_to_volume(
+    audio_file_url: str,
+) -> tuple[ParakeetUniqFilename, AudioFileExtension]:
+    import requests
+    from fastapi import HTTPException
+
+    response = requests.head(audio_file_url, allow_redirects=True)
+    if response.status_code == 404:
+        raise HTTPException(status_code=404, detail="Audio file not found")
+
+    response = requests.get(audio_file_url, allow_redirects=True)
+    response.raise_for_status()
+
+    audio_suffix = detect_audio_format(audio_file_url, response.headers)
+    unique_filename = ParakeetUniqFilename(f"{uuid.uuid4()}.{audio_suffix}")
+    file_path = f"{UPLOADS_PATH}/{unique_filename}"
+
+    with open(file_path, "wb") as f:
+        f.write(response.content)
+
+    upload_volume.commit()
+    return unique_filename, audio_suffix
+
+
+def pad_audio(audio_array, sample_rate: int = SAMPLERATE):
+    """Add 0.5 seconds of silence if audio is less than 500ms.
+
+    This is a workaround for a Parakeet bug where very short audio (<500ms) causes:
+    ValueError: `char_offsets`: [] and `processed_tokens`: [157, 834, 834, 841]
+    have to be of the same length
+
+    See: https://github.com/NVIDIA/NeMo/issues/8451
+    """
+    import numpy as np
+
+    audio_duration = len(audio_array) / sample_rate
+    if audio_duration < 0.5:
+        silence_samples = int(sample_rate * 0.5)
+        silence = np.zeros(silence_samples, dtype=np.float32)
+        return np.concatenate([audio_array, silence])
+    return audio_array
+
+
+@app.cls(
+    gpu="A10G",
+    timeout=600,
+    scaledown_window=300,
+    image=image,
+    volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
+    enable_memory_snapshot=True,
+    experimental_options={"enable_gpu_snapshot": True},
+)
+@modal.concurrent(max_inputs=10)
+class TranscriberParakeetLive:
+    @modal.enter(snap=True)
+    def enter(self):
+        import nemo.collections.asr as nemo_asr
+
+        logging.getLogger("nemo_logger").setLevel(logging.CRITICAL)
+
+        self.lock = threading.Lock()
+        self.model = nemo_asr.models.ASRModel.from_pretrained(model_name=MODEL_NAME)
+        device = next(self.model.parameters()).device
+        print(f"Model is on device: {device}")
+
+    @modal.method()
+    def transcribe_segment(
+        self,
+        filename: str,
+    ):
+        import librosa
+
+        upload_volume.reload()
+
+        file_path = f"{UPLOADS_PATH}/{filename}"
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(f"File not found: {file_path}")
+
+        audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
+        padded_audio = pad_audio(audio_array, sample_rate)
+
+        with self.lock:
+            with NoStdStreams():
+                (output,) = self.model.transcribe([padded_audio], timestamps=True)
+
+        text = output.text.strip()
+        words = [
+            {
+                "word": word_info["word"] + " ",
+                "start": round(word_info["start"], 2),
+                "end": round(word_info["end"], 2),
+            }
+            for word_info in output.timestamp["word"]
+        ]
+
+        return {"text": text, "words": words}
+
+    @modal.method()
+    def transcribe_batch(
+        self,
+        filenames: list[str],
+    ):
+        import librosa
+
+        upload_volume.reload()
+
+        results = []
+        audio_arrays = []
+
+        # Load all audio files with padding
+        for filename in filenames:
+            file_path = f"{UPLOADS_PATH}/{filename}"
+            if not os.path.exists(file_path):
+                raise FileNotFoundError(f"Batch file not found: {file_path}")
+
+            audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
+            padded_audio = pad_audio(audio_array, sample_rate)
+            audio_arrays.append(padded_audio)
+
+        with self.lock:
+            with NoStdStreams():
+                outputs = self.model.transcribe(audio_arrays, timestamps=True)
+
+        # Process results for each file
+        for i, (filename, output) in enumerate(zip(filenames, outputs)):
+            text = output.text.strip()
+
+            words = [
+                {
+                    "word": word_info["word"] + " ",
+                    "start": round(word_info["start"], 2),
+                    "end": round(word_info["end"], 2),
+                }
+                for word_info in output.timestamp["word"]
+            ]
+
+            results.append(
+                {
+                    "filename": filename,
+                    "text": text,
+                    "words": words,
+                }
+            )
+
+        return results
+
+
+# L40S class for file transcription (bigger files)
+@app.cls(
+    gpu="L40S",
+    timeout=900,
+    image=image,
+    volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
+    enable_memory_snapshot=True,
+    experimental_options={"enable_gpu_snapshot": True},
+)
+class TranscriberParakeetFile:
+    @modal.enter(snap=True)
+    def enter(self):
+        import nemo.collections.asr as nemo_asr
+        import torch
+        from silero_vad import load_silero_vad
+
+        logging.getLogger("nemo_logger").setLevel(logging.CRITICAL)
+
+        self.model = nemo_asr.models.ASRModel.from_pretrained(model_name=MODEL_NAME)
+        device = next(self.model.parameters()).device
+        print(f"Model is on device: {device}")
+
+        torch.set_num_threads(1)
+        self.vad_model = load_silero_vad(onnx=False)
+        print("Silero VAD initialized")
+
+    @modal.method()
+    def transcribe_segment(
+        self,
+        filename: str,
+        timestamp_offset: float = 0.0,
+    ):
+        import librosa
+        import numpy as np
+        from silero_vad import VADIterator
+
+        def load_and_convert_audio(file_path):
+            audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
+            return audio_array
+
+        def vad_segment_generator(audio_array):
+            """Generate speech segments using VAD with start/end sample indices"""
+            vad_iterator = VADIterator(self.vad_model, sampling_rate=SAMPLERATE)
+            window_size = VAD_CONFIG["window_size"]
+            start = None
+
+            for i in range(0, len(audio_array), window_size):
+                chunk = audio_array[i : i + window_size]
+                if len(chunk) < window_size:
+                    chunk = np.pad(
+                        chunk, (0, window_size - len(chunk)), mode="constant"
+                    )
+
+                speech_dict = vad_iterator(chunk)
+                if not speech_dict:
+                    continue
+
+                if "start" in speech_dict:
+                    start = speech_dict["start"]
+                    continue
+
+                if "end" in speech_dict and start is not None:
+                    end = speech_dict["end"]
+                    start_time = start / float(SAMPLERATE)
+                    end_time = end / float(SAMPLERATE)
+
+                    # Extract the actual audio segment
+                    audio_segment = audio_array[start:end]
+
+                    yield (start_time, end_time, audio_segment)
+                    start = None
+
+            vad_iterator.reset_states()
+
+        def vad_segment_filter(segments):
+            """Filter VAD segments by duration and chunk large segments"""
+            min_dur = VAD_CONFIG["min_segment_duration"]
+            max_dur = VAD_CONFIG["max_segment_duration"]
+
+            for start_time, end_time, audio_segment in segments:
+                segment_duration = end_time - start_time
+
+                # Skip very small segments
+                if segment_duration < min_dur:
+                    continue
+
+                # If segment is within max duration, yield as-is
+                if segment_duration <= max_dur:
+                    yield (start_time, end_time, audio_segment)
+                    continue
+
+                # Chunk large segments into smaller pieces
+                chunk_samples = int(max_dur * SAMPLERATE)
+                current_start = start_time
+
+                for chunk_offset in range(0, len(audio_segment), chunk_samples):
+                    chunk_audio = audio_segment[
+                        chunk_offset : chunk_offset + chunk_samples
+                    ]
+                    if len(chunk_audio) == 0:
+                        break
+
+                    chunk_duration = len(chunk_audio) / float(SAMPLERATE)
+                    chunk_end = current_start + chunk_duration
+
+                    # Only yield chunks that meet minimum duration
+                    if chunk_duration >= min_dur:
+                        yield (current_start, chunk_end, chunk_audio)
+
+                    current_start = chunk_end
+
+        def batch_segments(segments, max_files=10, max_duration=5.0):
+            batch = []
+            batch_duration = 0.0
+
+            for start_time, end_time, audio_segment in segments:
+                segment_duration = end_time - start_time
+
+                if segment_duration < VAD_CONFIG["silence_padding"]:
+                    silence_samples = int(
+                        (VAD_CONFIG["silence_padding"] - segment_duration) * SAMPLERATE
+                    )
+                    padding = np.zeros(silence_samples, dtype=np.float32)
+                    audio_segment = np.concatenate([audio_segment, padding])
+                    segment_duration = VAD_CONFIG["silence_padding"]
+
+                batch.append((start_time, end_time, audio_segment))
+                batch_duration += segment_duration
+
+                if len(batch) >= max_files or batch_duration >= max_duration:
+                    yield batch
+                    batch = []
+                    batch_duration = 0.0
+
+            if batch:
+                yield batch
+
+        def transcribe_batch(model, audio_segments):
+            with NoStdStreams():
+                outputs = model.transcribe(audio_segments, timestamps=True)
+            return outputs
+
+        def emit_results(
+            results,
+            segments_info,
+            batch_index,
+            total_batches,
+        ):
+            """Yield transcribed text and word timings from model output, adjusting timestamps to absolute positions."""
+            for i, (output, (start_time, end_time, _)) in enumerate(
+                zip(results, segments_info)
+            ):
+                text = output.text.strip()
+                words = [
+                    {
+                        "word": word_info["word"] + " ",
+                        "start": round(
+                            word_info["start"] + start_time + timestamp_offset, 2
+                        ),
+                        "end": round(
+                            word_info["end"] + start_time + timestamp_offset, 2
+                        ),
+                    }
+                    for word_info in output.timestamp["word"]
+                ]
+
+                yield text, words
+
+        upload_volume.reload()
+
+        file_path = f"{UPLOADS_PATH}/{filename}"
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(f"File not found: {file_path}")
+
+        audio_array = load_and_convert_audio(file_path)
+        total_duration = len(audio_array) / float(SAMPLERATE)
+        processed_duration = 0.0
+
+        all_text_parts = []
+        all_words = []
+
+        raw_segments = vad_segment_generator(audio_array)
+        filtered_segments = vad_segment_filter(raw_segments)
+        batches = batch_segments(
+            filtered_segments,
+            VAD_CONFIG["batch_max_files"],
+            VAD_CONFIG["batch_max_duration"],
+        )
+
+        batch_index = 0
+        total_batches = max(
+            1, int(total_duration / VAD_CONFIG["batch_max_duration"]) + 1
+        )
+
+        for batch in batches:
+            batch_index += 1
+            audio_segments = [seg[2] for seg in batch]
+            results = transcribe_batch(self.model, audio_segments)
+
+            for text, words in emit_results(
+                results,
+                batch,
+                batch_index,
+                total_batches,
+            ):
+                if not text:
+                    continue
+                all_text_parts.append(text)
+                all_words.extend(words)
+
+            processed_duration += sum(len(seg[2]) / float(SAMPLERATE) for seg in batch)
+
+        combined_text = " ".join(all_text_parts)
+        return {"text": combined_text, "words": all_words}
+
+
+@app.function(
+    scaledown_window=60,
+    timeout=600,
+    secrets=[
+        modal.Secret.from_name("reflector-gpu"),
+    ],
+    volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
+    image=image,
+)
+@modal.concurrent(max_inputs=40)
+@modal.asgi_app()
+def web():
+    import os
+    import uuid
+
+    from fastapi import (
+        Body,
+        Depends,
+        FastAPI,
+        Form,
+        HTTPException,
+        UploadFile,
+        status,
+    )
+    from fastapi.security import OAuth2PasswordBearer
+    from pydantic import BaseModel
+
+    transcriber_live = TranscriberParakeetLive()
+    transcriber_file = TranscriberParakeetFile()
+
+    app = FastAPI()
+
+    oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
+
+    def apikey_auth(apikey: str = Depends(oauth2_scheme)):
+        if apikey == os.environ["REFLECTOR_GPU_APIKEY"]:
+            return
+        raise HTTPException(
+            status_code=status.HTTP_401_UNAUTHORIZED,
+            detail="Invalid API key",
+            headers={"WWW-Authenticate": "Bearer"},
+        )
+
+    class TranscriptResponse(BaseModel):
+        result: dict
+
+    @app.post("/v1/audio/transcriptions", dependencies=[Depends(apikey_auth)])
+    def transcribe(
+        file: UploadFile = None,
+        files: list[UploadFile] | None = None,
+        model: str = Form(MODEL_NAME),
+        language: str = Form("en"),
+        batch: bool = Form(False),
+    ):
+        # Parakeet only supports English
+        if language != "en":
+            raise HTTPException(
+                status_code=400,
+                detail=f"Parakeet model only supports English. Got language='{language}'",
+            )
+        # Handle both single file and multiple files
+        if not file and not files:
+            raise HTTPException(
+                status_code=400, detail="Either 'file' or 'files' parameter is required"
+            )
+        if batch and not files:
+            raise HTTPException(
+                status_code=400, detail="Batch transcription requires 'files'"
+            )
+
+        upload_files = [file] if file else files
+
+        # Upload files to volume
+        uploaded_filenames = []
+        for upload_file in upload_files:
+            audio_suffix = upload_file.filename.split(".")[-1]
+            assert audio_suffix in SUPPORTED_FILE_EXTENSIONS
+
+            # Generate unique filename
+            unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
+            file_path = f"{UPLOADS_PATH}/{unique_filename}"
+
+            print(f"Writing file to: {file_path}")
+            with open(file_path, "wb") as f:
+                content = upload_file.file.read()
+                f.write(content)
+
+            uploaded_filenames.append(unique_filename)
+
+        upload_volume.commit()
+
+        try:
+            # Use A10G live transcriber for per-file transcription
+            if batch and len(upload_files) > 1:
+                # Use batch transcription
+                func = transcriber_live.transcribe_batch.spawn(
+                    filenames=uploaded_filenames,
+                )
+                results = func.get()
+                return {"results": results}
+
+            # Per-file transcription
+            results = []
+            for filename in uploaded_filenames:
+                func = transcriber_live.transcribe_segment.spawn(
+                    filename=filename,
+                )
+                result = func.get()
+                result["filename"] = filename
+                results.append(result)
+
+            return {"results": results} if len(results) > 1 else results[0]
+
+        finally:
+            for filename in uploaded_filenames:
+                try:
+                    file_path = f"{UPLOADS_PATH}/{filename}"
+                    print(f"Deleting file: {file_path}")
+                    os.remove(file_path)
+                except Exception as e:
+                    print(f"Error deleting {filename}: {e}")
+
+            upload_volume.commit()
+
+    @app.post("/v1/audio/transcriptions-from-url", dependencies=[Depends(apikey_auth)])
+    def transcribe_from_url(
+        audio_file_url: str = Body(
+            ..., description="URL of the audio file to transcribe"
+        ),
+        model: str = Body(MODEL_NAME),
+        language: str = Body("en", description="Language code (only 'en' supported)"),
+        timestamp_offset: float = Body(0.0),
+    ):
+        # Parakeet only supports English
+        if language != "en":
+            raise HTTPException(
+                status_code=400,
+                detail=f"Parakeet model only supports English. Got language='{language}'",
+            )
+        unique_filename, audio_suffix = download_audio_to_volume(audio_file_url)
+
+        try:
+            func = transcriber_file.transcribe_segment.spawn(
+                filename=unique_filename,
+                timestamp_offset=timestamp_offset,
+            )
+            result = func.get()
+            return result
+        finally:
+            try:
+                file_path = f"{UPLOADS_PATH}/{unique_filename}"
+                print(f"Deleting file: {file_path}")
+                os.remove(file_path)
+                upload_volume.commit()
+            except Exception as e:
+                print(f"Error cleaning up {unique_filename}: {e}")
+
+    return app
+
+
+class NoStdStreams:
+    def __init__(self):
+        self.devnull = open(os.devnull, "w")
+
+    def __enter__(self):
+        self._stdout, self._stderr = sys.stdout, sys.stderr
+        self._stdout.flush()
+        self._stderr.flush()
+        sys.stdout, sys.stderr = self.devnull, self.devnull
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        sys.stdout, sys.stderr = self._stdout, self._stderr
+        self.devnull.close()

server/migrations/README

@@ -1 +1,3 @@
 Generic single-database configuration.
+
+Both data migrations and schema migrations must be in migrations.

server/migrations/versions/0ab2d7ffaa16_add_long_summary_to_search_vector.py

@@ -0,0 +1,64 @@
+"""add_long_summary_to_search_vector
+
+Revision ID: 0ab2d7ffaa16
+Revises: b1c33bd09963
+Create Date: 2025-08-15 13:27:52.680211
+
+"""
+
+from typing import Sequence, Union
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "0ab2d7ffaa16"
+down_revision: Union[str, None] = "b1c33bd09963"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Drop the existing search vector column and index
+    op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
+    op.drop_column("transcript", "search_vector_en")
+
+    # Recreate the search vector column with long_summary included
+    op.execute("""
+        ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
+        GENERATED ALWAYS AS (
+            setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
+            setweight(to_tsvector('english', coalesce(long_summary, '')), 'B') ||
+            setweight(to_tsvector('english', coalesce(webvtt, '')), 'C')
+        ) STORED
+    """)
+
+    # Recreate the GIN index for the search vector
+    op.create_index(
+        "idx_transcript_search_vector_en",
+        "transcript",
+        ["search_vector_en"],
+        postgresql_using="gin",
+    )
+
+
+def downgrade() -> None:
+    # Drop the updated search vector column and index
+    op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
+    op.drop_column("transcript", "search_vector_en")
+
+    # Recreate the original search vector column without long_summary
+    op.execute("""
+        ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
+        GENERATED ALWAYS AS (
+            setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
+            setweight(to_tsvector('english', coalesce(webvtt, '')), 'B')
+        ) STORED
+    """)
+
+    # Recreate the GIN index for the search vector
+    op.create_index(
+        "idx_transcript_search_vector_en",
+        "transcript",
+        ["search_vector_en"],
+        postgresql_using="gin",
+    )

server/migrations/versions/0bc0f3ff0111_add_webvtt_field_to_transcript.py

@@ -0,0 +1,25 @@
+"""add_webvtt_field_to_transcript
+
+Revision ID: 0bc0f3ff0111
+Revises: b7df9609542c
+Create Date: 2025-08-05 19:36:41.740957
+
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+revision: str = "0bc0f3ff0111"
+down_revision: Union[str, None] = "b7df9609542c"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.add_column("transcript", sa.Column("webvtt", sa.Text(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("transcript", "webvtt")

server/migrations/versions/116b2f287eab_add_full_text_search.py

@@ -0,0 +1,46 @@
+"""add_full_text_search
+
+Revision ID: 116b2f287eab
+Revises: 0bc0f3ff0111
+Create Date: 2025-08-07 11:27:38.473517
+
+"""
+
+from typing import Sequence, Union
+
+from alembic import op
+
+revision: str = "116b2f287eab"
+down_revision: Union[str, None] = "0bc0f3ff0111"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name != "postgresql":
+        return
+
+    op.execute("""
+        ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
+        GENERATED ALWAYS AS (
+            setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
+            setweight(to_tsvector('english', coalesce(webvtt, '')), 'B')
+        ) STORED
+    """)
+
+    op.create_index(
+        "idx_transcript_search_vector_en",
+        "transcript",
+        ["search_vector_en"],
+        postgresql_using="gin",
+    )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    if conn.dialect.name != "postgresql":
+        return
+
+    op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
+    op.drop_column("transcript", "search_vector_en")

server/migrations/versions/62dea3db63a5_add_room_options.py

@@ -32,7 +32,7 @@ def upgrade() -> None:
         sa.Column("user_id", sa.String(), nullable=True),
         sa.Column("room_id", sa.String(), nullable=True),
         sa.Column(
-            "is_locked", sa.Boolean(), server_default=sa.text("0"), nullable=False
+            "is_locked", sa.Boolean(), server_default=sa.text("false"), nullable=False
         ),
         sa.Column("room_mode", sa.String(), server_default="normal", nullable=False),
         sa.Column(
@@ -53,12 +53,15 @@ def upgrade() -> None:
         sa.Column("user_id", sa.String(), nullable=False),
         sa.Column("created_at", sa.DateTime(), nullable=False),
         sa.Column(
-            "zulip_auto_post", sa.Boolean(), server_default=sa.text("0"), nullable=False
+            "zulip_auto_post",
+            sa.Boolean(),
+            server_default=sa.text("false"),
+            nullable=False,
         ),
         sa.Column("zulip_stream", sa.String(), nullable=True),
         sa.Column("zulip_topic", sa.String(), nullable=True),
         sa.Column(
-            "is_locked", sa.Boolean(), server_default=sa.text("0"), nullable=False
+            "is_locked", sa.Boolean(), server_default=sa.text("false"), nullable=False
         ),
         sa.Column("room_mode", sa.String(), server_default="normal", nullable=False),
         sa.Column(

server/migrations/versions/74b2b0236931_add_transcript_source_kind.py

@@ -20,11 +20,14 @@ depends_on: Union[str, Sequence[str], None] = None
 
 def upgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
+    sourcekind_enum = sa.Enum("room", "live", "file", name="sourcekind")
+    sourcekind_enum.create(op.get_bind())
+
     op.add_column(
         "transcript",
         sa.Column(
             "source_kind",
-            sa.Enum("ROOM", "LIVE", "FILE", name="sourcekind"),
+            sourcekind_enum,
             nullable=True,
         ),
     )
@@ -43,6 +46,8 @@ def upgrade() -> None:
 def downgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
     op.drop_column("transcript", "source_kind")
+    sourcekind_enum = sa.Enum(name="sourcekind")
+    sourcekind_enum.drop(op.get_bind())
 
 
 # ### end Alembic commands ###

server/migrations/versions/8120ebc75366_populate_webvtt_from_topics.py

@@ -0,0 +1,106 @@
+"""populate_webvtt_from_topics
+
+Revision ID: 8120ebc75366
+Revises: 116b2f287eab
+Create Date: 2025-08-11 19:11:01.316947
+
+"""
+
+import json
+from typing import Sequence, Union
+
+from alembic import op
+from sqlalchemy import text
+
+# revision identifiers, used by Alembic.
+revision: str = "8120ebc75366"
+down_revision: Union[str, None] = "116b2f287eab"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def topics_to_webvtt(topics):
+    """Convert topics list to WebVTT format string."""
+    if not topics:
+        return None
+
+    lines = ["WEBVTT", ""]
+
+    for topic in topics:
+        start_time = format_timestamp(topic.get("start"))
+        end_time = format_timestamp(topic.get("end"))
+        text = topic.get("text", "").strip()
+
+        if start_time and end_time and text:
+            lines.append(f"{start_time} --> {end_time}")
+            lines.append(text)
+            lines.append("")
+
+    return "\n".join(lines).strip()
+
+
+def format_timestamp(seconds):
+    """Format seconds to WebVTT timestamp format (HH:MM:SS.mmm)."""
+    if seconds is None:
+        return None
+
+    hours = int(seconds // 3600)
+    minutes = int((seconds % 3600) // 60)
+    secs = seconds % 60
+
+    return f"{hours:02d}:{minutes:02d}:{secs:06.3f}"
+
+
+def upgrade() -> None:
+    """Populate WebVTT field for all transcripts with topics."""
+
+    # Get connection
+    connection = op.get_bind()
+
+    # Query all transcripts with topics
+    result = connection.execute(
+        text("SELECT id, topics FROM transcript WHERE topics IS NOT NULL")
+    )
+
+    rows = result.fetchall()
+    print(f"Found {len(rows)} transcripts with topics")
+
+    updated_count = 0
+    error_count = 0
+
+    for row in rows:
+        transcript_id = row[0]
+        topics_data = row[1]
+
+        if not topics_data:
+            continue
+
+        try:
+            # Parse JSON if it's a string
+            if isinstance(topics_data, str):
+                topics_data = json.loads(topics_data)
+
+            # Convert topics to WebVTT format
+            webvtt_content = topics_to_webvtt(topics_data)
+
+            if webvtt_content:
+                # Update the webvtt field
+                connection.execute(
+                    text("UPDATE transcript SET webvtt = :webvtt WHERE id = :id"),
+                    {"webvtt": webvtt_content, "id": transcript_id},
+                )
+                updated_count += 1
+                print(f"✓ Updated transcript {transcript_id}")
+
+        except Exception as e:
+            error_count += 1
+            print(f"✗ Error updating transcript {transcript_id}: {e}")
+
+    print(f"\nMigration complete!")
+    print(f"  Updated: {updated_count}")
+    print(f"  Errors: {error_count}")
+
+
+def downgrade() -> None:
+    """Clear WebVTT field for all transcripts."""
+    op.execute(text("UPDATE transcript SET webvtt = NULL"))

server/migrations/versions/99365b0cd87b_add_title_short_and_long_summary_and_.py

@@ -22,7 +22,7 @@ def upgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
     op.execute(
         "UPDATE transcript SET events = "
-        'REPLACE(events, \'"event": "SUMMARY"\', \'"event": "LONG_SUMMARY"\');'
+        'REPLACE(events::text, \'"event": "SUMMARY"\', \'"event": "LONG_SUMMARY"\')::json;'
     )
     op.alter_column("transcript", "summary", new_column_name="long_summary")
     op.add_column("transcript", sa.Column("title", sa.String(), nullable=True))
@@ -34,7 +34,7 @@ def downgrade() -> None:
     # ### commands auto generated by Alembic - please adjust! ###
     op.execute(
         "UPDATE transcript SET events = "
-        'REPLACE(events, \'"event": "LONG_SUMMARY"\', \'"event": "SUMMARY"\');'
+        'REPLACE(events::text, \'"event": "LONG_SUMMARY"\', \'"event": "SUMMARY"\')::json;'
     )
     with op.batch_alter_table("transcript", schema=None) as batch_op:
         batch_op.alter_column("long_summary", nullable=True, new_column_name="summary")

server/migrations/versions/9f5c78d352d6_datetime_timezone.py

@@ -0,0 +1,121 @@
+"""datetime timezone
+
+Revision ID: 9f5c78d352d6
+Revises: 8120ebc75366
+Create Date: 2025-08-13 19:18:27.113593
+
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+from sqlalchemy.dialects import postgresql
+
+# revision identifiers, used by Alembic.
+revision: str = "9f5c78d352d6"
+down_revision: Union[str, None] = "8120ebc75366"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("meeting", schema=None) as batch_op:
+        batch_op.alter_column(
+            "start_date",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=True,
+        )
+        batch_op.alter_column(
+            "end_date",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=True,
+        )
+
+    with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
+        batch_op.alter_column(
+            "consent_timestamp",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("recording", schema=None) as batch_op:
+        batch_op.alter_column(
+            "recorded_at",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("room", schema=None) as batch_op:
+        batch_op.alter_column(
+            "created_at",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("transcript", schema=None) as batch_op:
+        batch_op.alter_column(
+            "created_at",
+            existing_type=postgresql.TIMESTAMP(),
+            type_=sa.DateTime(timezone=True),
+            existing_nullable=True,
+        )
+
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("transcript", schema=None) as batch_op:
+        batch_op.alter_column(
+            "created_at",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=True,
+        )
+
+    with op.batch_alter_table("room", schema=None) as batch_op:
+        batch_op.alter_column(
+            "created_at",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("recording", schema=None) as batch_op:
+        batch_op.alter_column(
+            "recorded_at",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
+        batch_op.alter_column(
+            "consent_timestamp",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=False,
+        )
+
+    with op.batch_alter_table("meeting", schema=None) as batch_op:
+        batch_op.alter_column(
+            "end_date",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=True,
+        )
+        batch_op.alter_column(
+            "start_date",
+            existing_type=sa.DateTime(timezone=True),
+            type_=postgresql.TIMESTAMP(),
+            existing_nullable=True,
+        )
+
+    # ### end Alembic commands ###

server/migrations/versions/a7122bc0b2ca_add_shared_rooms.py

@@ -25,7 +25,7 @@ def upgrade() -> None:
         sa.Column(
             "is_shared",
             sa.Boolean(),
-            server_default=sa.text("0"),
+            server_default=sa.text("false"),
             nullable=False,
         ),
     )

server/migrations/versions/b0e5f7876032_add_meeting_is_active.py

@@ -23,7 +23,10 @@ def upgrade() -> None:
     with op.batch_alter_table("meeting", schema=None) as batch_op:
         batch_op.add_column(
             sa.Column(
-                "is_active", sa.Boolean(), server_default=sa.text("1"), nullable=False
+                "is_active",
+                sa.Boolean(),
+                server_default=sa.text("true"),
+                nullable=False,
             )
         )
 

server/migrations/versions/b1c33bd09963_add_search_optimization_indexes.py

@@ -0,0 +1,41 @@
+"""add_search_optimization_indexes
+
+Revision ID: b1c33bd09963
+Revises: 9f5c78d352d6
+Create Date: 2025-08-14 17:26:02.117408
+
+"""
+
+from typing import Sequence, Union
+
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "b1c33bd09963"
+down_revision: Union[str, None] = "9f5c78d352d6"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Add indexes for actual search filtering patterns used in frontend
+    # Based on /browse page filters: room_id and source_kind
+
+    # Index for room_id + created_at (for room-specific searches with date ordering)
+    op.create_index(
+        "idx_transcript_room_id_created_at",
+        "transcript",
+        ["room_id", "created_at"],
+        if_not_exists=True,
+    )
+
+    # Index for source_kind alone (actively used filter in frontend)
+    op.create_index(
+        "idx_transcript_source_kind", "transcript", ["source_kind"], if_not_exists=True
+    )
+
+
+def downgrade() -> None:
+    # Remove the indexes in reverse order
+    op.drop_index("idx_transcript_source_kind", "transcript", if_exists=True)
+    op.drop_index("idx_transcript_room_id_created_at", "transcript", if_exists=True)

server/migrations/versions/b9348748bbbc_reviewed.py

@@ -23,7 +23,7 @@ def upgrade() -> None:
     op.add_column(
         "transcript",
         sa.Column(
-            "reviewed", sa.Boolean(), server_default=sa.text("0"), nullable=False
+            "reviewed", sa.Boolean(), server_default=sa.text("false"), nullable=False
         ),
     )
     # ### end Alembic commands ###

server/pyproject.toml

@@ -32,15 +32,14 @@ dependencies = [
     "redis>=5.0.1",
     "python-jose[cryptography]>=3.3.0",
     "python-multipart>=0.0.6",
-    "faster-whisper>=0.10.0",
     "transformers>=4.36.2",
-    "black==24.1.1",
     "jsonschema>=4.23.0",
     "openai>=1.59.7",
     "psycopg2-binary>=2.9.10",
     "llama-index>=0.12.52",
     "llama-index-llms-openai-like>=0.4.0",
     "pytest-env>=1.1.5",
+    "webvtt-py>=0.5.0",
 ]
 
 [dependency-groups]
@@ -57,6 +56,9 @@ tests = [
     "httpx-ws>=0.4.1",
     "pytest-httpx>=0.23.1",
     "pytest-celery>=0.0.0",
+    "pytest-recording>=0.13.4",
+    "pytest-docker>=3.2.3",
+    "asgi-lifespan>=2.1.0",
 ]
 aws = ["aioboto3>=11.2.0"]
 evaluation = [
@@ -65,6 +67,15 @@ evaluation = [
     "tqdm>=4.66.0",
     "pydantic>=2.1.1",
 ]
+local = [
+    "pyannote-audio>=3.3.2",
+    "faster-whisper>=0.10.0",
+]
+silero-vad = [
+    "silero-vad>=5.1.2",
+    "torch>=2.8.0",
+    "torchaudio>=2.8.0",
+]
 
 [tool.uv]
 default-groups = [
@@ -72,6 +83,21 @@ default-groups = [
     "tests",
     "aws",
     "evaluation",
+    "local",
+    "silero-vad"
+]
+
+[[tool.uv.index]]
+name = "pytorch-cpu"
+url = "https://download.pytorch.org/whl/cpu"
+explicit = true
+
+[tool.uv.sources]
+torch = [
+  { index = "pytorch-cpu" },
+]
+torchaudio = [
+  { index = "pytorch-cpu" },
 ]
 
 [build-system]
@@ -86,12 +112,26 @@ source = ["reflector"]
 
 [tool.pytest_env]
 ENVIRONMENT = "pytest"
-DATABASE_URL = "sqlite:///test.sqlite"
+DATABASE_URL = "postgresql://test_user:test_password@localhost:15432/reflector_test"
 
 [tool.pytest.ini_options]
 addopts = "-ra -q --disable-pytest-warnings --cov --cov-report html -v"
 testpaths = ["tests"]
 asyncio_mode = "auto"
+markers = [
+    "gpu_modal: mark test to run only with GPU Modal endpoints (deselect with '-m \"not gpu_modal\"')",
+]
+
+[tool.ruff.lint]
+select = [
+    "I",       # isort - import sorting
+    "F401",    # unused imports
+    "PLC0415", # import-outside-top-level - detect inline imports
+]
 
 [tool.ruff.lint.per-file-ignores]
 "reflector/processors/summary/summary_builder.py" = ["E501"]
+"gpu/**.py" = ["PLC0415"]
+"reflector/tools/**.py" = ["PLC0415"]
+"migrations/versions/**.py" = ["PLC0415"]
+"tests/**.py" = ["PLC0415"]

server/reflector/db/__init__.py

@@ -1,12 +1,28 @@
+import contextvars
+from typing import Optional
+
 import databases
 import sqlalchemy
 
 from reflector.events import subscribers_shutdown, subscribers_startup
 from reflector.settings import settings
 
-database = databases.Database(settings.DATABASE_URL)
 metadata = sqlalchemy.MetaData()
 
+_database_context: contextvars.ContextVar[Optional[databases.Database]] = (
+    contextvars.ContextVar("database", default=None)
+)
+
+
+def get_database() -> databases.Database:
+    """Get database instance for current asyncio context"""
+    db = _database_context.get()
+    if db is None:
+        db = databases.Database(settings.DATABASE_URL)
+        _database_context.set(db)
+    return db
+
+
 # import models
 import reflector.db.meetings  # noqa
 import reflector.db.recordings  # noqa
@@ -14,16 +30,18 @@ import reflector.db.rooms  # noqa
 import reflector.db.transcripts  # noqa
 
 kwargs = {}
-if "sqlite" in settings.DATABASE_URL:
-    kwargs["connect_args"] = {"check_same_thread": False}
+if "postgres" not in settings.DATABASE_URL:
+    raise Exception("Only postgres database is supported in reflector")
 engine = sqlalchemy.create_engine(settings.DATABASE_URL, **kwargs)
 
 
 @subscribers_startup.append
 async def database_connect(_):
+    database = get_database()
     await database.connect()
 
 
 @subscribers_shutdown.append
 async def database_disconnect(_):
+    database = get_database()
     await database.disconnect()

server/reflector/db/meetings.py

@@ -5,7 +5,7 @@ import sqlalchemy as sa
 from fastapi import HTTPException
 from pydantic import BaseModel, Field
 
-from reflector.db import database, metadata
+from reflector.db import get_database, metadata
 from reflector.db.rooms import Room
 from reflector.utils import generate_uuid4
 
@@ -16,8 +16,8 @@ meetings = sa.Table(
     sa.Column("room_name", sa.String),
     sa.Column("room_url", sa.String),
     sa.Column("host_room_url", sa.String),
-    sa.Column("start_date", sa.DateTime),
-    sa.Column("end_date", sa.DateTime),
+    sa.Column("start_date", sa.DateTime(timezone=True)),
+    sa.Column("end_date", sa.DateTime(timezone=True)),
     sa.Column("user_id", sa.String),
     sa.Column("room_id", sa.String),
     sa.Column("is_locked", sa.Boolean, nullable=False, server_default=sa.false()),
@@ -42,6 +42,12 @@ meetings = sa.Table(
         server_default=sa.true(),
     ),
     sa.Index("idx_meeting_room_id", "room_id"),
+    sa.Index(
+        "idx_one_active_meeting_per_room",
+        "room_id",
+        unique=True,
+        postgresql_where=sa.text("is_active = true"),
+    ),
 )
 
 meeting_consent = sa.Table(
@@ -51,7 +57,7 @@ meeting_consent = sa.Table(
     sa.Column("meeting_id", sa.String, sa.ForeignKey("meeting.id"), nullable=False),
     sa.Column("user_id", sa.String),
     sa.Column("consent_given", sa.Boolean, nullable=False),
-    sa.Column("consent_timestamp", sa.DateTime, nullable=False),
+    sa.Column("consent_timestamp", sa.DateTime(timezone=True), nullable=False),
 )
 
 
@@ -111,7 +117,7 @@ class MeetingController:
             recording_trigger=room.recording_trigger,
         )
         query = meetings.insert().values(**meeting.model_dump())
-        await database.execute(query)
+        await get_database().execute(query)
         return meeting
 
     async def get_all_active(self) -> list[Meeting]:
@@ -119,7 +125,7 @@ class MeetingController:
         Get active meetings.
         """
         query = meetings.select().where(meetings.c.is_active)
-        return await database.fetch_all(query)
+        return await get_database().fetch_all(query)
 
     async def get_by_room_name(
         self,
@@ -129,7 +135,7 @@ class MeetingController:
         Get a meeting by room name.
         """
         query = meetings.select().where(meetings.c.room_name == room_name)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
 
@@ -151,7 +157,7 @@ class MeetingController:
             )
             .order_by(end_date.desc())
         )
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
 
@@ -162,7 +168,7 @@ class MeetingController:
         Get a meeting by id
         """
         query = meetings.select().where(meetings.c.id == meeting_id)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
         return Meeting(**result)
@@ -174,7 +180,7 @@ class MeetingController:
         If not found, it will raise a 404 error.
         """
         query = meetings.select().where(meetings.c.id == meeting_id)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             raise HTTPException(status_code=404, detail="Meeting not found")
 
@@ -186,7 +192,7 @@ class MeetingController:
 
     async def update_meeting(self, meeting_id: str, **kwargs):
         query = meetings.update().where(meetings.c.id == meeting_id).values(**kwargs)
-        await database.execute(query)
+        await get_database().execute(query)
 
 
 class MeetingConsentController:
@@ -194,7 +200,7 @@ class MeetingConsentController:
         query = meeting_consent.select().where(
             meeting_consent.c.meeting_id == meeting_id
         )
-        results = await database.fetch_all(query)
+        results = await get_database().fetch_all(query)
         return [MeetingConsent(**result) for result in results]
 
     async def get_by_meeting_and_user(
@@ -205,7 +211,7 @@ class MeetingConsentController:
             meeting_consent.c.meeting_id == meeting_id,
             meeting_consent.c.user_id == user_id,
         )
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if result is None:
             return None
         return MeetingConsent(**result) if result else None
@@ -227,14 +233,14 @@ class MeetingConsentController:
                         consent_timestamp=consent.consent_timestamp,
                     )
                 )
-                await database.execute(query)
+                await get_database().execute(query)
 
                 existing.consent_given = consent.consent_given
                 existing.consent_timestamp = consent.consent_timestamp
                 return existing
 
         query = meeting_consent.insert().values(**consent.model_dump())
-        await database.execute(query)
+        await get_database().execute(query)
         return consent
 
     async def has_any_denial(self, meeting_id: str) -> bool:
@@ -243,7 +249,7 @@ class MeetingConsentController:
             meeting_consent.c.meeting_id == meeting_id,
             meeting_consent.c.consent_given.is_(False),
         )
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         return result is not None
 
 

server/reflector/db/recordings.py

@@ -4,7 +4,7 @@ from typing import Literal
 import sqlalchemy as sa
 from pydantic import BaseModel, Field
 
-from reflector.db import database, metadata
+from reflector.db import get_database, metadata
 from reflector.utils import generate_uuid4
 
 recordings = sa.Table(
@@ -13,7 +13,7 @@ recordings = sa.Table(
     sa.Column("id", sa.String, primary_key=True),
     sa.Column("bucket_name", sa.String, nullable=False),
     sa.Column("object_key", sa.String, nullable=False),
-    sa.Column("recorded_at", sa.DateTime, nullable=False),
+    sa.Column("recorded_at", sa.DateTime(timezone=True), nullable=False),
     sa.Column(
         "status",
         sa.String,
@@ -37,12 +37,12 @@ class Recording(BaseModel):
 class RecordingController:
     async def create(self, recording: Recording):
         query = recordings.insert().values(**recording.model_dump())
-        await database.execute(query)
+        await get_database().execute(query)
         return recording
 
     async def get_by_id(self, id: str) -> Recording:
         query = recordings.select().where(recordings.c.id == id)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         return Recording(**result) if result else None
 
     async def get_by_object_key(self, bucket_name: str, object_key: str) -> Recording:
@@ -50,8 +50,12 @@ class RecordingController:
             recordings.c.bucket_name == bucket_name,
             recordings.c.object_key == object_key,
         )
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         return Recording(**result) if result else None
 
+    async def remove_by_id(self, id: str) -> None:
+        query = recordings.delete().where(recordings.c.id == id)
+        await get_database().execute(query)
+
 
 recordings_controller = RecordingController()

server/reflector/db/rooms.py

@@ -1,4 +1,4 @@
-from datetime import datetime
+from datetime import datetime, timezone
 from sqlite3 import IntegrityError
 from typing import Literal
 
@@ -7,7 +7,7 @@ from fastapi import HTTPException
 from pydantic import BaseModel, Field
 from sqlalchemy.sql import false, or_
 
-from reflector.db import database, metadata
+from reflector.db import get_database, metadata
 from reflector.utils import generate_uuid4
 
 rooms = sqlalchemy.Table(
@@ -16,7 +16,7 @@ rooms = sqlalchemy.Table(
     sqlalchemy.Column("id", sqlalchemy.String, primary_key=True),
     sqlalchemy.Column("name", sqlalchemy.String, nullable=False, unique=True),
     sqlalchemy.Column("user_id", sqlalchemy.String, nullable=False),
-    sqlalchemy.Column("created_at", sqlalchemy.DateTime, nullable=False),
+    sqlalchemy.Column("created_at", sqlalchemy.DateTime(timezone=True), nullable=False),
     sqlalchemy.Column(
         "zulip_auto_post", sqlalchemy.Boolean, nullable=False, server_default=false()
     ),
@@ -48,7 +48,7 @@ class Room(BaseModel):
     id: str = Field(default_factory=generate_uuid4)
     name: str
     user_id: str
-    created_at: datetime = Field(default_factory=datetime.utcnow)
+    created_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
     zulip_auto_post: bool = False
     zulip_stream: str = ""
     zulip_topic: str = ""
@@ -92,7 +92,7 @@ class RoomController:
         if return_query:
             return query
 
-        results = await database.fetch_all(query)
+        results = await get_database().fetch_all(query)
         return results
 
     async def add(
@@ -125,7 +125,7 @@ class RoomController:
         )
         query = rooms.insert().values(**room.model_dump())
         try:
-            await database.execute(query)
+            await get_database().execute(query)
         except IntegrityError:
             raise HTTPException(status_code=400, detail="Room name is not unique")
         return room
@@ -136,7 +136,7 @@ class RoomController:
         """
         query = rooms.update().where(rooms.c.id == room.id).values(**values)
         try:
-            await database.execute(query)
+            await get_database().execute(query)
         except IntegrityError:
             raise HTTPException(status_code=400, detail="Room name is not unique")
 
@@ -151,7 +151,7 @@ class RoomController:
         query = rooms.select().where(rooms.c.id == room_id)
         if "user_id" in kwargs:
             query = query.where(rooms.c.user_id == kwargs["user_id"])
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
         return Room(**result)
@@ -163,7 +163,7 @@ class RoomController:
         query = rooms.select().where(rooms.c.name == room_name)
         if "user_id" in kwargs:
             query = query.where(rooms.c.user_id == kwargs["user_id"])
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
         return Room(**result)
@@ -175,7 +175,7 @@ class RoomController:
         If not found, it will raise a 404 error.
         """
         query = rooms.select().where(rooms.c.id == meeting_id)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             raise HTTPException(status_code=404, detail="Room not found")
 
@@ -197,7 +197,7 @@ class RoomController:
         if user_id is not None and room.user_id != user_id:
             return
         query = rooms.delete().where(rooms.c.id == room_id)
-        await database.execute(query)
+        await get_database().execute(query)
 
 
 rooms_controller = RoomController()

server/reflector/db/search.py

@@ -0,0 +1,448 @@
+"""Search functionality for transcripts and other entities."""
+
+import itertools
+from dataclasses import dataclass
+from datetime import datetime
+from io import StringIO
+from typing import Annotated, Any, Dict, Iterator
+
+import sqlalchemy
+import webvtt
+from fastapi import HTTPException
+from pydantic import (
+    BaseModel,
+    Field,
+    NonNegativeFloat,
+    NonNegativeInt,
+    ValidationError,
+    constr,
+    field_serializer,
+)
+
+from reflector.db import get_database
+from reflector.db.rooms import rooms
+from reflector.db.transcripts import SourceKind, transcripts
+from reflector.db.utils import is_postgresql
+from reflector.logger import logger
+
+DEFAULT_SEARCH_LIMIT = 20
+SNIPPET_CONTEXT_LENGTH = 50  # Characters before/after match to include
+DEFAULT_SNIPPET_MAX_LENGTH = NonNegativeInt(150)
+DEFAULT_MAX_SNIPPETS = NonNegativeInt(3)
+LONG_SUMMARY_MAX_SNIPPETS = 2
+
+SearchQueryBase = constr(min_length=0, strip_whitespace=True)
+SearchLimitBase = Annotated[int, Field(ge=1, le=100)]
+SearchOffsetBase = Annotated[int, Field(ge=0)]
+SearchTotalBase = Annotated[int, Field(ge=0)]
+
+SearchQuery = Annotated[SearchQueryBase, Field(description="Search query text")]
+SearchLimit = Annotated[SearchLimitBase, Field(description="Results per page")]
+SearchOffset = Annotated[
+    SearchOffsetBase, Field(description="Number of results to skip")
+]
+SearchTotal = Annotated[
+    SearchTotalBase, Field(description="Total number of search results")
+]
+
+WEBVTT_SPEC_HEADER = "WEBVTT"
+
+WebVTTContent = Annotated[
+    str,
+    Field(min_length=len(WEBVTT_SPEC_HEADER), description="WebVTT content"),
+]
+
+
+class WebVTTProcessor:
+    """Stateless processor for WebVTT content operations."""
+
+    @staticmethod
+    def parse(raw_content: str) -> WebVTTContent:
+        """Parse WebVTT content and return it as a string."""
+        if not raw_content.startswith(WEBVTT_SPEC_HEADER):
+            raise ValueError(f"Invalid WebVTT content, no header {WEBVTT_SPEC_HEADER}")
+        return raw_content
+
+    @staticmethod
+    def extract_text(webvtt_content: WebVTTContent) -> str:
+        """Extract plain text from WebVTT content using webvtt library."""
+        try:
+            buffer = StringIO(webvtt_content)
+            vtt = webvtt.read_buffer(buffer)
+            return " ".join(caption.text for caption in vtt if caption.text)
+        except webvtt.errors.MalformedFileError as e:
+            logger.warning(f"Malformed WebVTT content: {e}")
+            return ""
+        except (UnicodeDecodeError, ValueError) as e:
+            logger.warning(f"Failed to decode WebVTT content: {e}")
+            return ""
+        except AttributeError as e:
+            logger.error(
+                f"WebVTT parsing error - unexpected format: {e}", exc_info=True
+            )
+            return ""
+        except Exception as e:
+            logger.error(f"Unexpected error parsing WebVTT: {e}", exc_info=True)
+            return ""
+
+    @staticmethod
+    def generate_snippets(
+        webvtt_content: WebVTTContent,
+        query: str,
+        max_snippets: NonNegativeInt = DEFAULT_MAX_SNIPPETS,
+    ) -> list[str]:
+        """Generate snippets from WebVTT content."""
+        return SnippetGenerator.generate(
+            WebVTTProcessor.extract_text(webvtt_content),
+            query,
+            max_snippets=max_snippets,
+        )
+
+
+@dataclass(frozen=True)
+class SnippetCandidate:
+    """Represents a candidate snippet with its position."""
+
+    _text: str
+    start: NonNegativeInt
+    _original_text_length: int
+
+    @property
+    def end(self) -> NonNegativeInt:
+        """Calculate end position from start and raw text length."""
+        return self.start + len(self._text)
+
+    def text(self) -> str:
+        """Get display text with ellipses added if needed."""
+        result = self._text.strip()
+        if self.start > 0:
+            result = "..." + result
+        if self.end < self._original_text_length:
+            result = result + "..."
+        return result
+
+
+class SearchParameters(BaseModel):
+    """Validated search parameters for full-text search."""
+
+    query_text: SearchQuery
+    limit: SearchLimit = DEFAULT_SEARCH_LIMIT
+    offset: SearchOffset = 0
+    user_id: str | None = None
+    room_id: str | None = None
+    source_kind: SourceKind | None = None
+
+
+class SearchResultDB(BaseModel):
+    """Intermediate model for validating raw database results."""
+
+    id: str = Field(..., min_length=1)
+    created_at: datetime
+    status: str = Field(..., min_length=1)
+    duration: float | None = Field(None, ge=0)
+    user_id: str | None = None
+    title: str | None = None
+    source_kind: SourceKind
+    room_id: str | None = None
+    rank: float = Field(..., ge=0, le=1)
+
+
+class SearchResult(BaseModel):
+    """Public search result model with computed fields."""
+
+    id: str = Field(..., min_length=1)
+    title: str | None = None
+    user_id: str | None = None
+    room_id: str | None = None
+    room_name: str | None = None
+    source_kind: SourceKind
+    created_at: datetime
+    status: str = Field(..., min_length=1)
+    rank: float = Field(..., ge=0, le=1)
+    duration: NonNegativeFloat | None = Field(..., description="Duration in seconds")
+    search_snippets: list[str] = Field(
+        description="Text snippets around search matches"
+    )
+    total_match_count: NonNegativeInt = Field(
+        default=0, description="Total number of matches found in the transcript"
+    )
+
+    @field_serializer("created_at", when_used="json")
+    def serialize_datetime(self, dt: datetime) -> str:
+        if dt.tzinfo is None:
+            return dt.isoformat() + "Z"
+        return dt.isoformat()
+
+
+class SnippetGenerator:
+    """Stateless generator for text snippets and match operations."""
+
+    @staticmethod
+    def find_all_matches(text: str, query: str) -> Iterator[int]:
+        """Generate all match positions for a query in text."""
+        if not text:
+            logger.warning("Empty text for search query in find_all_matches")
+            return
+        if not query:
+            logger.warning("Empty query for search text in find_all_matches")
+            return
+
+        text_lower = text.lower()
+        query_lower = query.lower()
+        start = 0
+        prev_start = start
+        while (pos := text_lower.find(query_lower, start)) != -1:
+            yield pos
+            start = pos + len(query_lower)
+            if start <= prev_start:
+                raise ValueError("panic! find_all_matches is not incremental")
+            prev_start = start
+
+    @staticmethod
+    def count_matches(text: str, query: str) -> NonNegativeInt:
+        """Count total number of matches for a query in text."""
+        ZERO = NonNegativeInt(0)
+        if not text:
+            logger.warning("Empty text for search query in count_matches")
+            return ZERO
+        if not query:
+            logger.warning("Empty query for search text in count_matches")
+            return ZERO
+        return NonNegativeInt(
+            sum(1 for _ in SnippetGenerator.find_all_matches(text, query))
+        )
+
+    @staticmethod
+    def create_snippet(
+        text: str, match_pos: int, max_length: int = DEFAULT_SNIPPET_MAX_LENGTH
+    ) -> SnippetCandidate:
+        """Create a snippet from a match position."""
+        snippet_start = NonNegativeInt(max(0, match_pos - SNIPPET_CONTEXT_LENGTH))
+        snippet_end = min(len(text), match_pos + max_length - SNIPPET_CONTEXT_LENGTH)
+
+        snippet_text = text[snippet_start:snippet_end]
+
+        return SnippetCandidate(
+            _text=snippet_text, start=snippet_start, _original_text_length=len(text)
+        )
+
+    @staticmethod
+    def filter_non_overlapping(
+        candidates: Iterator[SnippetCandidate],
+    ) -> Iterator[str]:
+        """Filter out overlapping snippets and return only display text."""
+        last_end = 0
+        for candidate in candidates:
+            display_text = candidate.text()
+            # it means that next overlapping snippets simply don't get included
+            # it's fine as simplistic logic and users probably won't care much because they already have their search results just fin
+            if candidate.start >= last_end and display_text:
+                yield display_text
+                last_end = candidate.end
+
+    @staticmethod
+    def generate(
+        text: str,
+        query: str,
+        max_length: NonNegativeInt = DEFAULT_SNIPPET_MAX_LENGTH,
+        max_snippets: NonNegativeInt = DEFAULT_MAX_SNIPPETS,
+    ) -> list[str]:
+        """Generate snippets from text."""
+        if not text or not query:
+            logger.warning("Empty text or query for generate_snippets")
+            return []
+
+        candidates = (
+            SnippetGenerator.create_snippet(text, pos, max_length)
+            for pos in SnippetGenerator.find_all_matches(text, query)
+        )
+        filtered = SnippetGenerator.filter_non_overlapping(candidates)
+        snippets = list(itertools.islice(filtered, max_snippets))
+
+        # Fallback to first word search if no full matches
+        # it's another assumption: proper snippet logic generation is quite complicated and tied to db logic, so simplification is used here
+        if not snippets and " " in query:
+            first_word = query.split()[0]
+            return SnippetGenerator.generate(text, first_word, max_length, max_snippets)
+
+        return snippets
+
+    @staticmethod
+    def from_summary(
+        summary: str,
+        query: str,
+        max_snippets: NonNegativeInt = LONG_SUMMARY_MAX_SNIPPETS,
+    ) -> list[str]:
+        """Generate snippets from summary text."""
+        return SnippetGenerator.generate(summary, query, max_snippets=max_snippets)
+
+    @staticmethod
+    def combine_sources(
+        summary: str | None,
+        webvtt: WebVTTContent | None,
+        query: str,
+        max_total: NonNegativeInt = DEFAULT_MAX_SNIPPETS,
+    ) -> tuple[list[str], NonNegativeInt]:
+        """Combine snippets from multiple sources and return total match count.
+
+        Returns (snippets, total_match_count) tuple.
+
+        snippets can be empty for real in case of e.g. title match
+        """
+        webvtt_matches = 0
+        summary_matches = 0
+
+        if webvtt:
+            webvtt_text = WebVTTProcessor.extract_text(webvtt)
+            webvtt_matches = SnippetGenerator.count_matches(webvtt_text, query)
+
+        if summary:
+            summary_matches = SnippetGenerator.count_matches(summary, query)
+
+        total_matches = NonNegativeInt(webvtt_matches + summary_matches)
+
+        summary_snippets = (
+            SnippetGenerator.from_summary(summary, query) if summary else []
+        )
+
+        if len(summary_snippets) >= max_total:
+            return summary_snippets[:max_total], total_matches
+
+        remaining = max_total - len(summary_snippets)
+        webvtt_snippets = (
+            WebVTTProcessor.generate_snippets(webvtt, query, remaining)
+            if webvtt
+            else []
+        )
+
+        return summary_snippets + webvtt_snippets, total_matches
+
+
+class SearchController:
+    """Controller for search operations across different entities."""
+
+    @classmethod
+    async def search_transcripts(
+        cls, params: SearchParameters
+    ) -> tuple[list[SearchResult], int]:
+        """
+        Full-text search for transcripts using PostgreSQL tsvector.
+        Returns (results, total_count).
+        """
+
+        if not is_postgresql():
+            logger.warning(
+                "Full-text search requires PostgreSQL. Returning empty results."
+            )
+            return [], 0
+
+        base_columns = [
+            transcripts.c.id,
+            transcripts.c.title,
+            transcripts.c.created_at,
+            transcripts.c.duration,
+            transcripts.c.status,
+            transcripts.c.user_id,
+            transcripts.c.room_id,
+            transcripts.c.source_kind,
+            transcripts.c.webvtt,
+            transcripts.c.long_summary,
+            sqlalchemy.case(
+                (
+                    transcripts.c.room_id.isnot(None) & rooms.c.id.is_(None),
+                    "Deleted Room",
+                ),
+                else_=rooms.c.name,
+            ).label("room_name"),
+        ]
+
+        if params.query_text:
+            search_query = sqlalchemy.func.websearch_to_tsquery(
+                "english", params.query_text
+            )
+            rank_column = sqlalchemy.func.ts_rank(
+                transcripts.c.search_vector_en,
+                search_query,
+                32,  # normalization flag: rank/(rank+1) for 0-1 range
+            ).label("rank")
+        else:
+            rank_column = sqlalchemy.cast(1.0, sqlalchemy.Float).label("rank")
+
+        columns = base_columns + [rank_column]
+        base_query = sqlalchemy.select(columns).select_from(
+            transcripts.join(rooms, transcripts.c.room_id == rooms.c.id, isouter=True)
+        )
+
+        if params.query_text:
+            base_query = base_query.where(
+                transcripts.c.search_vector_en.op("@@")(search_query)
+            )
+
+        if params.user_id:
+            base_query = base_query.where(
+                sqlalchemy.or_(
+                    transcripts.c.user_id == params.user_id, rooms.c.is_shared
+                )
+            )
+        else:
+            base_query = base_query.where(rooms.c.is_shared)
+        if params.room_id:
+            base_query = base_query.where(transcripts.c.room_id == params.room_id)
+        if params.source_kind:
+            base_query = base_query.where(
+                transcripts.c.source_kind == params.source_kind
+            )
+
+        if params.query_text:
+            order_by = sqlalchemy.desc(sqlalchemy.text("rank"))
+        else:
+            order_by = sqlalchemy.desc(transcripts.c.created_at)
+
+        query = base_query.order_by(order_by).limit(params.limit).offset(params.offset)
+
+        rs = await get_database().fetch_all(query)
+
+        count_query = sqlalchemy.select([sqlalchemy.func.count()]).select_from(
+            base_query.alias("search_results")
+        )
+        total = await get_database().fetch_val(count_query)
+
+        def _process_result(r) -> SearchResult:
+            r_dict: Dict[str, Any] = dict(r)
+            webvtt_raw: str | None = r_dict.pop("webvtt", None)
+            if webvtt_raw:
+                webvtt = WebVTTProcessor.parse(webvtt_raw)
+            else:
+                webvtt = None
+            long_summary: str | None = r_dict.pop("long_summary", None)
+            room_name: str | None = r_dict.pop("room_name", None)
+            db_result = SearchResultDB.model_validate(r_dict)
+
+            snippets, total_match_count = SnippetGenerator.combine_sources(
+                long_summary, webvtt, params.query_text, DEFAULT_MAX_SNIPPETS
+            )
+
+            return SearchResult(
+                **db_result.model_dump(),
+                room_name=room_name,
+                search_snippets=snippets,
+                total_match_count=total_match_count,
+            )
+
+        try:
+            results = [_process_result(r) for r in rs]
+        except ValidationError as e:
+            logger.error(f"Invalid search result data: {e}", exc_info=True)
+            raise HTTPException(
+                status_code=500, detail="Internal search result data consistency error"
+            )
+        except Exception as e:
+            logger.error(f"Error processing search results: {e}", exc_info=True)
+            raise
+
+        return results, total
+
+
+search_controller = SearchController()
+webvtt_processor = WebVTTProcessor()
+snippet_generator = SnippetGenerator()

server/reflector/db/transcripts.py

@@ -3,7 +3,7 @@ import json
 import os
 import shutil
 from contextlib import asynccontextmanager
-from datetime import datetime, timezone
+from datetime import datetime, timedelta, timezone
 from pathlib import Path
 from typing import Any, Literal
 
@@ -11,13 +11,19 @@ import sqlalchemy
 from fastapi import HTTPException
 from pydantic import BaseModel, ConfigDict, Field, field_serializer
 from sqlalchemy import Enum
+from sqlalchemy.dialects.postgresql import TSVECTOR
 from sqlalchemy.sql import false, or_
 
-from reflector.db import database, metadata
+from reflector.db import get_database, metadata
+from reflector.db.recordings import recordings_controller
+from reflector.db.rooms import rooms
+from reflector.db.utils import is_postgresql
+from reflector.logger import logger
 from reflector.processors.types import Word as ProcessorWord
 from reflector.settings import settings
-from reflector.storage import get_transcripts_storage
+from reflector.storage import get_recordings_storage, get_transcripts_storage
 from reflector.utils import generate_uuid4
+from reflector.utils.webvtt import topics_to_webvtt
 
 
 class SourceKind(enum.StrEnum):
@@ -34,7 +40,7 @@ transcripts = sqlalchemy.Table(
     sqlalchemy.Column("status", sqlalchemy.String),
     sqlalchemy.Column("locked", sqlalchemy.Boolean),
     sqlalchemy.Column("duration", sqlalchemy.Float),
-    sqlalchemy.Column("created_at", sqlalchemy.DateTime),
+    sqlalchemy.Column("created_at", sqlalchemy.DateTime(timezone=True)),
     sqlalchemy.Column("title", sqlalchemy.String),
     sqlalchemy.Column("short_summary", sqlalchemy.String),
     sqlalchemy.Column("long_summary", sqlalchemy.String),
@@ -76,13 +82,40 @@ transcripts = sqlalchemy.Table(
     # same field could've been in recording/meeting, and it's maybe even ok to dupe it at need
     sqlalchemy.Column("audio_deleted", sqlalchemy.Boolean),
     sqlalchemy.Column("room_id", sqlalchemy.String),
+    sqlalchemy.Column("webvtt", sqlalchemy.Text),
     sqlalchemy.Index("idx_transcript_recording_id", "recording_id"),
     sqlalchemy.Index("idx_transcript_user_id", "user_id"),
     sqlalchemy.Index("idx_transcript_created_at", "created_at"),
     sqlalchemy.Index("idx_transcript_user_id_recording_id", "user_id", "recording_id"),
     sqlalchemy.Index("idx_transcript_room_id", "room_id"),
+    sqlalchemy.Index("idx_transcript_source_kind", "source_kind"),
+    sqlalchemy.Index("idx_transcript_room_id_created_at", "room_id", "created_at"),
 )
 
+# Add PostgreSQL-specific full-text search column
+# This matches the migration in migrations/versions/116b2f287eab_add_full_text_search.py
+if is_postgresql():
+    transcripts.append_column(
+        sqlalchemy.Column(
+            "search_vector_en",
+            TSVECTOR,
+            sqlalchemy.Computed(
+                "setweight(to_tsvector('english', coalesce(title, '')), 'A') || "
+                "setweight(to_tsvector('english', coalesce(long_summary, '')), 'B') || "
+                "setweight(to_tsvector('english', coalesce(webvtt, '')), 'C')",
+                persisted=True,
+            ),
+        )
+    )
+    # Add GIN index for the search vector
+    transcripts.append_constraint(
+        sqlalchemy.Index(
+            "idx_transcript_search_vector_en",
+            "search_vector_en",
+            postgresql_using="gin",
+        )
+    )
+
 
 def generate_transcript_name() -> str:
     now = datetime.now(timezone.utc)
@@ -147,14 +180,18 @@ class TranscriptParticipant(BaseModel):
 
 
 class Transcript(BaseModel):
+    """Full transcript model with all fields."""
+
     id: str = Field(default_factory=generate_uuid4)
     user_id: str | None = None
     name: str = Field(default_factory=generate_transcript_name)
     status: str = "idle"
-    locked: bool = False
     duration: float = 0
     created_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
     title: str | None = None
+    source_kind: SourceKind
+    room_id: str | None = None
+    locked: bool = False
     short_summary: str | None = None
     long_summary: str | None = None
     topics: list[TranscriptTopic] = []
@@ -168,9 +205,8 @@ class Transcript(BaseModel):
     meeting_id: str | None = None
     recording_id: str | None = None
     zulip_message_id: int | None = None
-    source_kind: SourceKind
     audio_deleted: bool | None = None
-    room_id: str | None = None
+    webvtt: str | None = None
 
     @field_serializer("created_at", when_used="json")
     def serialize_datetime(self, dt: datetime) -> str:
@@ -271,10 +307,12 @@ class Transcript(BaseModel):
         # we need to create an url to be used for diarization
         # we can't use the audio_mp3_filename because it's not accessible
         # from the diarization processor
-        from datetime import timedelta
 
-        from reflector.app import app
-        from reflector.views.transcripts import create_access_token
+        # TODO don't import app in db
+        from reflector.app import app  # noqa: PLC0415
+
+        # TODO a util + don''t import views in db
+        from reflector.views.transcripts import create_access_token  # noqa: PLC0415
 
         path = app.url_path_for(
             "transcript_get_audio_mp3",
@@ -335,7 +373,6 @@ class TranscriptController:
         - `room_id`: filter transcripts by room ID
         - `search_term`: filter transcripts by search term
         """
-        from reflector.db.rooms import rooms
 
         query = transcripts.select().join(
             rooms, transcripts.c.room_id == rooms.c.id, isouter=True
@@ -386,7 +423,7 @@ class TranscriptController:
         if return_query:
             return query
 
-        results = await database.fetch_all(query)
+        results = await get_database().fetch_all(query)
         return results
 
     async def get_by_id(self, transcript_id: str, **kwargs) -> Transcript | None:
@@ -396,7 +433,7 @@ class TranscriptController:
         query = transcripts.select().where(transcripts.c.id == transcript_id)
         if "user_id" in kwargs:
             query = query.where(transcripts.c.user_id == kwargs["user_id"])
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
         return Transcript(**result)
@@ -410,7 +447,7 @@ class TranscriptController:
         query = transcripts.select().where(transcripts.c.recording_id == recording_id)
         if "user_id" in kwargs:
             query = query.where(transcripts.c.user_id == kwargs["user_id"])
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             return None
         return Transcript(**result)
@@ -428,7 +465,7 @@ class TranscriptController:
             if order_by.startswith("-"):
                 field = field.desc()
             query = query.order_by(field)
-        results = await database.fetch_all(query)
+        results = await get_database().fetch_all(query)
         return [Transcript(**result) for result in results]
 
     async def get_by_id_for_http(
@@ -446,7 +483,7 @@ class TranscriptController:
         to determine if the user can access the transcript.
         """
         query = transcripts.select().where(transcripts.c.id == transcript_id)
-        result = await database.fetch_one(query)
+        result = await get_database().fetch_one(query)
         if not result:
             raise HTTPException(status_code=404, detail="Transcript not found")
 
@@ -499,23 +536,52 @@ class TranscriptController:
             room_id=room_id,
         )
         query = transcripts.insert().values(**transcript.model_dump())
-        await database.execute(query)
+        await get_database().execute(query)
         return transcript
 
-    async def update(self, transcript: Transcript, values: dict, mutate=True):
+    # TODO investigate why mutate= is used. it's used in one place currently, maybe because of ORM field updates.
+    # using mutate=True is discouraged
+    async def update(
+        self, transcript: Transcript, values: dict, mutate=False
+    ) -> Transcript:
         """
-        Update a transcript fields with key/values in values
+        Update a transcript fields with key/values in values.
+        Returns a copy of the transcript with updated values.
         """
+        values = TranscriptController._handle_topics_update(values)
+
         query = (
             transcripts.update()
             .where(transcripts.c.id == transcript.id)
             .values(**values)
         )
-        await database.execute(query)
+        await get_database().execute(query)
         if mutate:
             for key, value in values.items():
                 setattr(transcript, key, value)
 
+        updated_transcript = transcript.model_copy(update=values)
+        return updated_transcript
+
+    @staticmethod
+    def _handle_topics_update(values: dict) -> dict:
+        """Auto-update WebVTT when topics are updated."""
+
+        if values.get("webvtt") is not None:
+            logger.warn("trying to update read-only webvtt column")
+            pass
+
+        topics_data = values.get("topics")
+        if topics_data is None:
+            return values
+
+        return {
+            **values,
+            "webvtt": topics_to_webvtt(
+                [TranscriptTopic(**topic_dict) for topic_dict in topics_data]
+            ),
+        }
+
     async def remove_by_id(
         self,
         transcript_id: str,
@@ -529,23 +595,55 @@ class TranscriptController:
             return
         if user_id is not None and transcript.user_id != user_id:
             return
+        if transcript.audio_location == "storage" and not transcript.audio_deleted:
+            try:
+                await get_transcripts_storage().delete_file(
+                    transcript.storage_audio_path
+                )
+            except Exception as e:
+                logger.warning(
+                    "Failed to delete transcript audio from storage",
+                    exc_info=e,
+                    transcript_id=transcript.id,
+                )
         transcript.unlink()
+        if transcript.recording_id:
+            try:
+                recording = await recordings_controller.get_by_id(
+                    transcript.recording_id
+                )
+                if recording:
+                    try:
+                        await get_recordings_storage().delete_file(recording.object_key)
+                    except Exception as e:
+                        logger.warning(
+                            "Failed to delete recording object from S3",
+                            exc_info=e,
+                            recording_id=transcript.recording_id,
+                        )
+                    await recordings_controller.remove_by_id(transcript.recording_id)
+            except Exception as e:
+                logger.warning(
+                    "Failed to delete recording row",
+                    exc_info=e,
+                    recording_id=transcript.recording_id,
+                )
         query = transcripts.delete().where(transcripts.c.id == transcript_id)
-        await database.execute(query)
+        await get_database().execute(query)
 
     async def remove_by_recording_id(self, recording_id: str):
         """
         Remove a transcript by recording_id
         """
         query = transcripts.delete().where(transcripts.c.recording_id == recording_id)
-        await database.execute(query)
+        await get_database().execute(query)
 
     @asynccontextmanager
     async def transaction(self):
         """
         A context manager for database transaction
         """
-        async with database.transaction(isolation="serializable"):
+        async with get_database().transaction(isolation="serializable"):
             yield
 
     async def append_event(
@@ -558,11 +656,7 @@ class TranscriptController:
         Append an event to a transcript
         """
         resp = transcript.add_event(event=event, data=data)
-        await self.update(
-            transcript,
-            {"events": transcript.events_dump()},
-            mutate=False,
-        )
+        await self.update(transcript, {"events": transcript.events_dump()})
         return resp
 
     async def upsert_topic(
@@ -574,11 +668,7 @@ class TranscriptController:
         Upsert topics to a transcript
         """
         transcript.upsert_topic(topic)
-        await self.update(
-            transcript,
-            {"topics": transcript.topics_dump()},
-            mutate=False,
-        )
+        await self.update(transcript, {"topics": transcript.topics_dump()})
 
     async def move_mp3_to_storage(self, transcript: Transcript):
         """
@@ -603,7 +693,8 @@ class TranscriptController:
             )
 
             # indicate on the transcript that the audio is now on storage
-            await self.update(transcript, {"audio_location": "storage"})
+            # mutates transcript argument
+            await self.update(transcript, {"audio_location": "storage"}, mutate=True)
 
         # unlink the local file
         transcript.audio_mp3_filename.unlink(missing_ok=True)
@@ -627,11 +718,7 @@ class TranscriptController:
         Add/update a participant to a transcript
         """
         result = transcript.upsert_participant(participant)
-        await self.update(
-            transcript,
-            {"participants": transcript.participants_dump()},
-            mutate=False,
-        )
+        await self.update(transcript, {"participants": transcript.participants_dump()})
         return result
 
     async def delete_participant(
@@ -643,11 +730,7 @@ class TranscriptController:
         Delete a participant from a transcript
         """
         transcript.delete_participant(participant_id)
-        await self.update(
-            transcript,
-            {"participants": transcript.participants_dump()},
-            mutate=False,
-        )
+        await self.update(transcript, {"participants": transcript.participants_dump()})
 
 
 transcripts_controller = TranscriptController()

server/reflector/db/utils.py

@@ -0,0 +1,9 @@
+"""Database utility functions."""
+
+from reflector.db import get_database
+
+
+def is_postgresql() -> bool:
+    return get_database().url.scheme and get_database().url.scheme.startswith(
+        "postgresql"
+    )

server/reflector/pipelines/main_file_pipeline.py

@@ -0,0 +1,375 @@
+"""
+File-based processing pipeline
+==============================
+
+Optimized pipeline for processing complete audio/video files.
+Uses parallel processing for transcription, diarization, and waveform generation.
+"""
+
+import asyncio
+from pathlib import Path
+
+import av
+import structlog
+from celery import shared_task
+
+from reflector.db.transcripts import (
+    Transcript,
+    transcripts_controller,
+)
+from reflector.logger import logger
+from reflector.pipelines.main_live_pipeline import PipelineMainBase, asynctask
+from reflector.processors import (
+    AudioFileWriterProcessor,
+    TranscriptFinalSummaryProcessor,
+    TranscriptFinalTitleProcessor,
+    TranscriptTopicDetectorProcessor,
+)
+from reflector.processors.audio_waveform_processor import AudioWaveformProcessor
+from reflector.processors.file_diarization import FileDiarizationInput
+from reflector.processors.file_diarization_auto import FileDiarizationAutoProcessor
+from reflector.processors.file_transcript import FileTranscriptInput
+from reflector.processors.file_transcript_auto import FileTranscriptAutoProcessor
+from reflector.processors.transcript_diarization_assembler import (
+    TranscriptDiarizationAssemblerInput,
+    TranscriptDiarizationAssemblerProcessor,
+)
+from reflector.processors.types import (
+    DiarizationSegment,
+    TitleSummary,
+)
+from reflector.processors.types import (
+    Transcript as TranscriptType,
+)
+from reflector.settings import settings
+from reflector.storage import get_transcripts_storage
+
+
+class EmptyPipeline:
+    """Empty pipeline for processors that need a pipeline reference"""
+
+    def __init__(self, logger: structlog.BoundLogger):
+        self.logger = logger
+
+    def get_pref(self, k, d=None):
+        return d
+
+    async def emit(self, event):
+        pass
+
+
+class PipelineMainFile(PipelineMainBase):
+    """
+    Optimized file processing pipeline.
+    Processes complete audio/video files with parallel execution.
+    """
+
+    logger: structlog.BoundLogger = None
+    empty_pipeline = None
+
+    def __init__(self, transcript_id: str):
+        super().__init__(transcript_id=transcript_id)
+        self.logger = logger.bind(transcript_id=self.transcript_id)
+        self.empty_pipeline = EmptyPipeline(logger=self.logger)
+
+    def _handle_gather_exceptions(self, results: list, operation: str) -> None:
+        """Handle exceptions from asyncio.gather with return_exceptions=True"""
+        for i, result in enumerate(results):
+            if not isinstance(result, Exception):
+                continue
+            self.logger.error(
+                f"Error in {operation} (task {i}): {result}",
+                transcript_id=self.transcript_id,
+                exc_info=result,
+            )
+
+    async def process(self, file_path: Path):
+        """Main entry point for file processing"""
+        self.logger.info(f"Starting file pipeline for {file_path}")
+
+        transcript = await self.get_transcript()
+
+        # Extract audio and write to transcript location
+        audio_path = await self.extract_and_write_audio(file_path, transcript)
+
+        # Upload for processing
+        audio_url = await self.upload_audio(audio_path, transcript)
+
+        # Run parallel processing
+        await self.run_parallel_processing(
+            audio_path,
+            audio_url,
+            transcript.source_language,
+            transcript.target_language,
+        )
+
+        self.logger.info("File pipeline complete")
+
+    async def extract_and_write_audio(
+        self, file_path: Path, transcript: Transcript
+    ) -> Path:
+        """Extract audio from video if needed and write to transcript location as MP3"""
+        self.logger.info(f"Processing audio file: {file_path}")
+
+        # Check if it's already audio-only
+        container = av.open(str(file_path))
+        has_video = len(container.streams.video) > 0
+        container.close()
+
+        # Use AudioFileWriterProcessor to write MP3 to transcript location
+        mp3_writer = AudioFileWriterProcessor(
+            path=transcript.audio_mp3_filename,
+            on_duration=self.on_duration,
+        )
+
+        # Process audio frames and write to transcript location
+        input_container = av.open(str(file_path))
+        for frame in input_container.decode(audio=0):
+            await mp3_writer.push(frame)
+
+        await mp3_writer.flush()
+        input_container.close()
+
+        if has_video:
+            self.logger.info(
+                f"Extracted audio from video and saved to {transcript.audio_mp3_filename}"
+            )
+        else:
+            self.logger.info(
+                f"Converted audio file and saved to {transcript.audio_mp3_filename}"
+            )
+
+        return transcript.audio_mp3_filename
+
+    async def upload_audio(self, audio_path: Path, transcript: Transcript) -> str:
+        """Upload audio to storage for processing"""
+        storage = get_transcripts_storage()
+
+        if not storage:
+            raise Exception(
+                "Storage backend required for file processing. Configure TRANSCRIPT_STORAGE_* settings."
+            )
+
+        self.logger.info("Uploading audio to storage")
+
+        with open(audio_path, "rb") as f:
+            audio_data = f.read()
+
+        storage_path = f"file_pipeline/{transcript.id}/audio.mp3"
+        await storage.put_file(storage_path, audio_data)
+
+        audio_url = await storage.get_file_url(storage_path)
+
+        self.logger.info(f"Audio uploaded to {audio_url}")
+        return audio_url
+
+    async def run_parallel_processing(
+        self,
+        audio_path: Path,
+        audio_url: str,
+        source_language: str,
+        target_language: str,
+    ):
+        """Coordinate parallel processing of transcription, diarization, and waveform"""
+        self.logger.info(
+            "Starting parallel processing", transcript_id=self.transcript_id
+        )
+
+        # Phase 1: Parallel processing of independent tasks
+        transcription_task = self.transcribe_file(audio_url, source_language)
+        diarization_task = self.diarize_file(audio_url)
+        waveform_task = self.generate_waveform(audio_path)
+
+        results = await asyncio.gather(
+            transcription_task, diarization_task, waveform_task, return_exceptions=True
+        )
+
+        transcript_result = results[0]
+        diarization_result = results[1]
+
+        # Handle errors - raise any exception that occurred
+        self._handle_gather_exceptions(results, "parallel processing")
+        for result in results:
+            if isinstance(result, Exception):
+                raise result
+
+        # Phase 2: Assemble transcript with diarization
+        self.logger.info(
+            "Assembling transcript with diarization", transcript_id=self.transcript_id
+        )
+        processor = TranscriptDiarizationAssemblerProcessor()
+        input_data = TranscriptDiarizationAssemblerInput(
+            transcript=transcript_result, diarization=diarization_result or []
+        )
+
+        # Store result for retrieval
+        diarized_transcript: Transcript | None = None
+
+        async def capture_result(transcript):
+            nonlocal diarized_transcript
+            diarized_transcript = transcript
+
+        processor.on(capture_result)
+        await processor.push(input_data)
+        await processor.flush()
+
+        if not diarized_transcript:
+            raise ValueError("No diarized transcript captured")
+
+        # Phase 3: Generate topics from diarized transcript
+        self.logger.info("Generating topics", transcript_id=self.transcript_id)
+        topics = await self.detect_topics(diarized_transcript, target_language)
+
+        # Phase 4: Generate title and summaries in parallel
+        self.logger.info(
+            "Generating title and summaries", transcript_id=self.transcript_id
+        )
+        results = await asyncio.gather(
+            self.generate_title(topics),
+            self.generate_summaries(topics),
+            return_exceptions=True,
+        )
+
+        self._handle_gather_exceptions(results, "title and summary generation")
+
+    async def transcribe_file(self, audio_url: str, language: str) -> TranscriptType:
+        """Transcribe complete file"""
+        processor = FileTranscriptAutoProcessor()
+        input_data = FileTranscriptInput(audio_url=audio_url, language=language)
+
+        # Store result for retrieval
+        result: TranscriptType | None = None
+
+        async def capture_result(transcript):
+            nonlocal result
+            result = transcript
+
+        processor.on(capture_result)
+        await processor.push(input_data)
+        await processor.flush()
+
+        if not result:
+            raise ValueError("No transcript captured")
+
+        return result
+
+    async def diarize_file(self, audio_url: str) -> list[DiarizationSegment] | None:
+        """Get diarization for file"""
+        if not settings.DIARIZATION_BACKEND:
+            self.logger.info("Diarization disabled")
+            return None
+
+        processor = FileDiarizationAutoProcessor()
+        input_data = FileDiarizationInput(audio_url=audio_url)
+
+        # Store result for retrieval
+        result = None
+
+        async def capture_result(diarization_output):
+            nonlocal result
+            result = diarization_output.diarization
+
+        try:
+            processor.on(capture_result)
+            await processor.push(input_data)
+            await processor.flush()
+            return result
+        except Exception as e:
+            self.logger.error(f"Diarization failed: {e}")
+            return None
+
+    async def generate_waveform(self, audio_path: Path):
+        """Generate and save waveform"""
+        transcript = await self.get_transcript()
+
+        processor = AudioWaveformProcessor(
+            audio_path=audio_path,
+            waveform_path=transcript.audio_waveform_filename,
+            on_waveform=self.on_waveform,
+        )
+        processor.set_pipeline(self.empty_pipeline)
+
+        await processor.flush()
+
+    async def detect_topics(
+        self, transcript: TranscriptType, target_language: str
+    ) -> list[TitleSummary]:
+        """Detect topics from complete transcript"""
+        chunk_size = 300
+        topics: list[TitleSummary] = []
+
+        async def on_topic(topic: TitleSummary):
+            topics.append(topic)
+            return await self.on_topic(topic)
+
+        topic_detector = TranscriptTopicDetectorProcessor(callback=on_topic)
+        topic_detector.set_pipeline(self.empty_pipeline)
+
+        for i in range(0, len(transcript.words), chunk_size):
+            chunk_words = transcript.words[i : i + chunk_size]
+            if not chunk_words:
+                continue
+
+            chunk_transcript = TranscriptType(
+                words=chunk_words, translation=transcript.translation
+            )
+
+            await topic_detector.push(chunk_transcript)
+
+        await topic_detector.flush()
+        return topics
+
+    async def generate_title(self, topics: list[TitleSummary]):
+        """Generate title from topics"""
+        if not topics:
+            self.logger.warning("No topics for title generation")
+            return
+
+        processor = TranscriptFinalTitleProcessor(callback=self.on_title)
+        processor.set_pipeline(self.empty_pipeline)
+
+        for topic in topics:
+            await processor.push(topic)
+
+        await processor.flush()
+
+    async def generate_summaries(self, topics: list[TitleSummary]):
+        """Generate long and short summaries from topics"""
+        if not topics:
+            self.logger.warning("No topics for summary generation")
+            return
+
+        transcript = await self.get_transcript()
+        processor = TranscriptFinalSummaryProcessor(
+            transcript=transcript,
+            callback=self.on_long_summary,
+            on_short_summary=self.on_short_summary,
+        )
+        processor.set_pipeline(self.empty_pipeline)
+
+        for topic in topics:
+            await processor.push(topic)
+
+        await processor.flush()
+
+
+@shared_task
+@asynctask
+async def task_pipeline_file_process(*, transcript_id: str):
+    """Celery task for file pipeline processing"""
+
+    transcript = await transcripts_controller.get_by_id(transcript_id)
+    if not transcript:
+        raise Exception(f"Transcript {transcript_id} not found")
+
+    # Find the file to process
+    audio_file = next(transcript.data_path.glob("upload.*"), None)
+    if not audio_file:
+        audio_file = next(transcript.data_path.glob("audio.*"), None)
+
+    if not audio_file:
+        raise Exception("No audio file found to process")
+
+    # Run file pipeline
+    pipeline = PipelineMainFile(transcript_id=transcript_id)
+    await pipeline.process(audio_file)

server/reflector/pipelines/main_live_pipeline.py

@@ -14,12 +14,15 @@ It is directly linked to our data model.
 import asyncio
 import functools
 from contextlib import asynccontextmanager
+from typing import Generic
 
+import av
 import boto3
 from celery import chord, current_task, group, shared_task
 from pydantic import BaseModel
 from structlog import BoundLogger as Logger
 
+from reflector.db import get_database
 from reflector.db.meetings import meeting_consent_controller, meetings_controller
 from reflector.db.recordings import recordings_controller
 from reflector.db.rooms import rooms_controller
@@ -35,10 +38,11 @@ from reflector.db.transcripts import (
     transcripts_controller,
 )
 from reflector.logger import logger
-from reflector.pipelines.runner import PipelineRunner
+from reflector.pipelines.runner import PipelineMessage, PipelineRunner
 from reflector.processors import (
-    AudioChunkerProcessor,
+    AudioChunkerAutoProcessor,
     AudioDiarizationAutoProcessor,
+    AudioDownscaleProcessor,
     AudioFileWriterProcessor,
     AudioMergeProcessor,
     AudioTranscriptAutoProcessor,
@@ -69,8 +73,7 @@ def asynctask(f):
     @functools.wraps(f)
     def wrapper(*args, **kwargs):
         async def run_with_db():
-            from reflector.db import database
-
+            database = get_database()
             await database.connect()
             try:
                 return await f(*args, **kwargs)
@@ -144,16 +147,19 @@ class StrValue(BaseModel):
     value: str
 
 
-class PipelineMainBase(PipelineRunner):
-    transcript_id: str
-    ws_room_id: str | None = None
-    ws_manager: WebsocketManager | None = None
-
-    def prepare(self):
-        # prepare websocket
+class PipelineMainBase(PipelineRunner[PipelineMessage], Generic[PipelineMessage]):
+    def __init__(self, transcript_id: str):
+        super().__init__()
         self._lock = asyncio.Lock()
+        self.transcript_id = transcript_id
         self.ws_room_id = f"ts:{self.transcript_id}"
-        self.ws_manager = get_ws_manager()
+        self._ws_manager = None
+
+    @property
+    def ws_manager(self) -> WebsocketManager:
+        if self._ws_manager is None:
+            self._ws_manager = get_ws_manager()
+        return self._ws_manager
 
     async def get_transcript(self) -> Transcript:
         # fetch the transcript
@@ -164,7 +170,11 @@ class PipelineMainBase(PipelineRunner):
             raise Exception("Transcript not found")
         return result
 
-    def get_transcript_topics(self, transcript: Transcript) -> list[TranscriptTopic]:
+    @staticmethod
+    def wrap_transcript_topics(
+        topics: list[TranscriptTopic],
+    ) -> list[TitleSummaryWithIdProcessorType]:
+        # transformation to a pipe-supported format
         return [
             TitleSummaryWithIdProcessorType(
                 id=topic.id,
@@ -174,7 +184,7 @@ class PipelineMainBase(PipelineRunner):
                 duration=topic.duration,
                 transcript=TranscriptProcessorType(words=topic.words),
             )
-            for topic in transcript.topics
+            for topic in topics
         ]
 
     @asynccontextmanager
@@ -349,7 +359,6 @@ class PipelineMainLive(PipelineMainBase):
     async def create(self) -> Pipeline:
         # create a context for the whole rtc transaction
         # add a customised logger to the context
-        self.prepare()
         transcript = await self.get_transcript()
 
         processors = [
@@ -357,7 +366,8 @@ class PipelineMainLive(PipelineMainBase):
                 path=transcript.audio_wav_filename,
                 on_duration=self.on_duration,
             ),
-            AudioChunkerProcessor(),
+            AudioDownscaleProcessor(),
+            AudioChunkerAutoProcessor(),
             AudioMergeProcessor(),
             AudioTranscriptAutoProcessor.as_threaded(),
             TranscriptLinerProcessor(),
@@ -370,6 +380,7 @@ class PipelineMainLive(PipelineMainBase):
         pipeline.set_pref("audio:target_language", transcript.target_language)
         pipeline.logger.bind(transcript_id=transcript.id)
         pipeline.logger.info("Pipeline main live created")
+        pipeline.describe()
 
         return pipeline
 
@@ -380,7 +391,7 @@ class PipelineMainLive(PipelineMainBase):
         pipeline_post(transcript_id=self.transcript_id)
 
 
-class PipelineMainDiarization(PipelineMainBase):
+class PipelineMainDiarization(PipelineMainBase[AudioDiarizationInput]):
     """
     Diarize the audio and update topics
     """
@@ -388,7 +399,6 @@ class PipelineMainDiarization(PipelineMainBase):
     async def create(self) -> Pipeline:
         # create a context for the whole rtc transaction
         # add a customised logger to the context
-        self.prepare()
         pipeline = Pipeline(
             AudioDiarizationAutoProcessor(callback=self.on_topic),
         )
@@ -404,11 +414,10 @@ class PipelineMainDiarization(PipelineMainBase):
             pipeline.logger.info("Audio is local, skipping diarization")
             return
 
-        topics = self.get_transcript_topics(transcript)
         audio_url = await transcript.get_audio_url()
         audio_diarization_input = AudioDiarizationInput(
             audio_url=audio_url,
-            topics=topics,
+            topics=self.wrap_transcript_topics(transcript.topics),
         )
 
         # as tempting to use pipeline.push, prefer to use the runner
@@ -421,7 +430,7 @@ class PipelineMainDiarization(PipelineMainBase):
         return pipeline
 
 
-class PipelineMainFromTopics(PipelineMainBase):
+class PipelineMainFromTopics(PipelineMainBase[TitleSummaryWithIdProcessorType]):
     """
     Pseudo class for generating a pipeline from topics
     """
@@ -430,8 +439,6 @@ class PipelineMainFromTopics(PipelineMainBase):
         raise NotImplementedError
 
     async def create(self) -> Pipeline:
-        self.prepare()
-
         # get transcript
         self._transcript = transcript = await self.get_transcript()
 
@@ -443,7 +450,7 @@ class PipelineMainFromTopics(PipelineMainBase):
         pipeline.logger.info(f"{self.__class__.__name__} pipeline created")
 
         # push topics
-        topics = self.get_transcript_topics(transcript)
+        topics = PipelineMainBase.wrap_transcript_topics(transcript.topics)
         for topic in topics:
             await self.push(topic)
 
@@ -524,8 +531,6 @@ async def pipeline_convert_to_mp3(transcript: Transcript, logger: Logger):
     # Convert to mp3
     mp3_filename = transcript.audio_mp3_filename
 
-    import av
-
     with av.open(wav_filename.as_posix()) as in_container:
         in_stream = in_container.streams.audio[0]
         with av.open(mp3_filename.as_posix(), "w") as out_container:
@@ -604,7 +609,7 @@ async def cleanup_consent(transcript: Transcript, logger: Logger):
                         meeting.id
                     )
     except Exception as e:
-        logger.error(f"Failed to get fetch consent: {e}")
+        logger.error(f"Failed to get fetch consent: {e}", exc_info=e)
         consent_denied = True
 
     if not consent_denied:
@@ -627,7 +632,7 @@ async def cleanup_consent(transcript: Transcript, logger: Logger):
                 f"Deleted original Whereby recording: {recording.bucket_name}/{recording.object_key}"
             )
         except Exception as e:
-            logger.error(f"Failed to delete Whereby recording: {e}")
+            logger.error(f"Failed to delete Whereby recording: {e}", exc_info=e)
 
     # non-transactional, files marked for deletion not actually deleted is possible
     await transcripts_controller.update(transcript, {"audio_deleted": True})
@@ -640,7 +645,7 @@ async def cleanup_consent(transcript: Transcript, logger: Logger):
                 f"Deleted processed audio from storage: {transcript.storage_audio_path}"
             )
         except Exception as e:
-            logger.error(f"Failed to delete processed audio: {e}")
+            logger.error(f"Failed to delete processed audio: {e}", exc_info=e)
 
     # 3. Delete local audio files
     try:
@@ -649,7 +654,7 @@ async def cleanup_consent(transcript: Transcript, logger: Logger):
         if hasattr(transcript, "audio_wav_filename") and transcript.audio_wav_filename:
             transcript.audio_wav_filename.unlink(missing_ok=True)
     except Exception as e:
-        logger.error(f"Failed to delete local audio files: {e}")
+        logger.error(f"Failed to delete local audio files: {e}", exc_info=e)
 
     logger.info("Consent cleanup done")
 
@@ -794,8 +799,6 @@ def pipeline_post(*, transcript_id: str):
 
 @get_transcript
 async def pipeline_process(transcript: Transcript, logger: Logger):
-    import av
-
     try:
         if transcript.audio_location == "storage":
             await transcripts_controller.download_mp3_from_storage(transcript)

server/reflector/pipelines/runner.py

@@ -16,21 +16,16 @@ During its lifecycle, it will emit the following status:
 """
 
 import asyncio
-
-from pydantic import BaseModel, ConfigDict
+from typing import Generic, TypeVar
 
 from reflector.logger import logger
 from reflector.processors import Pipeline
 
+PipelineMessage = TypeVar("PipelineMessage")
 
-class PipelineRunner(BaseModel):
-    model_config = ConfigDict(arbitrary_types_allowed=True)
 
-    status: str = "idle"
-    pipeline: Pipeline | None = None
-
-    def __init__(self, **kwargs):
-        super().__init__(**kwargs)
+class PipelineRunner(Generic[PipelineMessage]):
+    def __init__(self):
         self._task = None
         self._q_cmd = asyncio.Queue(maxsize=4096)
         self._ev_done = asyncio.Event()
@@ -39,6 +34,8 @@ class PipelineRunner(BaseModel):
             runner=id(self),
             runner_cls=self.__class__.__name__,
         )
+        self.status = "idle"
+        self.pipeline: Pipeline | None = None
 
     async def create(self) -> Pipeline:
         """
@@ -67,7 +64,7 @@ class PipelineRunner(BaseModel):
         coro = self.run()
         asyncio.run(coro)
 
-    async def push(self, data):
+    async def push(self, data: PipelineMessage):
         """
         Push data to the pipeline
         """
@@ -92,7 +89,11 @@ class PipelineRunner(BaseModel):
         pass
 
     async def _add_cmd(
-        self, cmd: str, data, max_retries: int = 3, retry_time_limit: int = 3
+        self,
+        cmd: str,
+        data: PipelineMessage,
+        max_retries: int = 3,
+        retry_time_limit: int = 3,
     ):
         """
         Enqueue a command to be executed in the runner.
@@ -143,6 +144,9 @@ class PipelineRunner(BaseModel):
                 cmd, data = await self._q_cmd.get()
                 func = getattr(self, f"cmd_{cmd.lower()}")
                 if func:
+                    if cmd.upper() == "FLUSH":
+                        await func()
+                    else:
                         await func(data)
                 else:
                     raise Exception(f"Unknown command {cmd}")
@@ -152,13 +156,13 @@ class PipelineRunner(BaseModel):
             self._ev_done.set()
             raise
 
-    async def cmd_push(self, data):
+    async def cmd_push(self, data: PipelineMessage):
         if self._is_first_push:
             await self._set_status("push")
             self._is_first_push = False
         await self.pipeline.push(data)
 
-    async def cmd_flush(self, data):
+    async def cmd_flush(self):
         await self._set_status("flush")
         await self.pipeline.flush()
         await self._set_status("ended")

server/reflector/processors/__init__.py

@@ -1,5 +1,7 @@
 from .audio_chunker import AudioChunkerProcessor  # noqa: F401
+from .audio_chunker_auto import AudioChunkerAutoProcessor  # noqa: F401
 from .audio_diarization_auto import AudioDiarizationAutoProcessor  # noqa: F401
+from .audio_downscale import AudioDownscaleProcessor  # noqa: F401
 from .audio_file_writer import AudioFileWriterProcessor  # noqa: F401
 from .audio_merge import AudioMergeProcessor  # noqa: F401
 from .audio_transcript import AudioTranscriptProcessor  # noqa: F401
@@ -11,6 +13,13 @@ from .base import (  # noqa: F401
     Processor,
     ThreadedProcessor,
 )
+from .file_diarization import FileDiarizationProcessor  # noqa: F401
+from .file_diarization_auto import FileDiarizationAutoProcessor  # noqa: F401
+from .file_transcript import FileTranscriptProcessor  # noqa: F401
+from .file_transcript_auto import FileTranscriptAutoProcessor  # noqa: F401
+from .transcript_diarization_assembler import (
+    TranscriptDiarizationAssemblerProcessor,  # noqa: F401
+)
 from .transcript_final_summary import TranscriptFinalSummaryProcessor  # noqa: F401
 from .transcript_final_title import TranscriptFinalTitleProcessor  # noqa: F401
 from .transcript_liner import TranscriptLinerProcessor  # noqa: F401

server/reflector/processors/audio_chunker.py

@@ -1,28 +1,78 @@
+from typing import Optional
+
 import av
+from prometheus_client import Counter, Histogram
 
 from reflector.processors.base import Processor
 
 
 class AudioChunkerProcessor(Processor):
     """
-    Assemble audio frames into chunks
+    Base class for assembling audio frames into chunks
     """
 
     INPUT_TYPE = av.AudioFrame
     OUTPUT_TYPE = list[av.AudioFrame]
 
-    def __init__(self, max_frames=256):
-        super().__init__()
+    m_chunk = Histogram(
+        "audio_chunker",
+        "Time spent in AudioChunker.chunk",
+        ["backend"],
+    )
+    m_chunk_call = Counter(
+        "audio_chunker_call",
+        "Number of calls to AudioChunker.chunk",
+        ["backend"],
+    )
+    m_chunk_success = Counter(
+        "audio_chunker_success",
+        "Number of successful calls to AudioChunker.chunk",
+        ["backend"],
+    )
+    m_chunk_failure = Counter(
+        "audio_chunker_failure",
+        "Number of failed calls to AudioChunker.chunk",
+        ["backend"],
+    )
+
+    def __init__(self, *args, **kwargs):
+        name = self.__class__.__name__
+        self.m_chunk = self.m_chunk.labels(name)
+        self.m_chunk_call = self.m_chunk_call.labels(name)
+        self.m_chunk_success = self.m_chunk_success.labels(name)
+        self.m_chunk_failure = self.m_chunk_failure.labels(name)
+        super().__init__(*args, **kwargs)
         self.frames: list[av.AudioFrame] = []
-        self.max_frames = max_frames
 
     async def _push(self, data: av.AudioFrame):
-        self.frames.append(data)
-        if len(self.frames) >= self.max_frames:
-            await self.flush()
+        """Process incoming audio frame"""
+        # Validate audio format on first frame
+        if len(self.frames) == 0:
+            if data.sample_rate != 16000 or len(data.layout.channels) != 1:
+                raise ValueError(
+                    f"AudioChunkerProcessor expects 16kHz mono audio, got {data.sample_rate}Hz "
+                    f"with {len(data.layout.channels)} channel(s). "
+                    f"Use AudioDownscaleProcessor before this processor."
+                )
+
+        try:
+            self.m_chunk_call.inc()
+            with self.m_chunk.time():
+                result = await self._chunk(data)
+            self.m_chunk_success.inc()
+            if result:
+                await self.emit(result)
+        except Exception:
+            self.m_chunk_failure.inc()
+            raise
+
+    async def _chunk(self, data: av.AudioFrame) -> Optional[list[av.AudioFrame]]:
+        """
+        Process audio frame and return chunk when ready.
+        Subclasses should implement their chunking logic here.
+        """
+        raise NotImplementedError
 
     async def _flush(self):
-        frames = self.frames[:]
-        self.frames = []
-        if frames:
-            await self.emit(frames)
+        """Flush any remaining frames when processing ends"""
+        raise NotImplementedError

server/reflector/processors/audio_chunker_auto.py

@@ -0,0 +1,32 @@
+import importlib
+
+from reflector.processors.audio_chunker import AudioChunkerProcessor
+from reflector.settings import settings
+
+
+class AudioChunkerAutoProcessor(AudioChunkerProcessor):
+    _registry = {}
+
+    @classmethod
+    def register(cls, name, kclass):
+        cls._registry[name] = kclass
+
+    def __new__(cls, name: str | None = None, **kwargs):
+        if name is None:
+            name = settings.AUDIO_CHUNKER_BACKEND
+        if name not in cls._registry:
+            module_name = f"reflector.processors.audio_chunker_{name}"
+            importlib.import_module(module_name)
+
+        # gather specific configuration for the processor
+        # search `AUDIO_CHUNKER_BACKEND_XXX_YYY`, push to constructor as `backend_xxx_yyy`
+        config = {}
+        name_upper = name.upper()
+        settings_prefix = "AUDIO_CHUNKER_"
+        config_prefix = f"{settings_prefix}{name_upper}_"
+        for key, value in settings:
+            if key.startswith(config_prefix):
+                config_name = key[len(settings_prefix) :].lower()
+                config[config_name] = value
+
+        return cls._registry[name](**config | kwargs)

server/reflector/processors/audio_chunker_frames.py

@@ -0,0 +1,34 @@
+from typing import Optional
+
+import av
+
+from reflector.processors.audio_chunker import AudioChunkerProcessor
+from reflector.processors.audio_chunker_auto import AudioChunkerAutoProcessor
+
+
+class AudioChunkerFramesProcessor(AudioChunkerProcessor):
+    """
+    Simple frame-based audio chunker that emits chunks after a fixed number of frames
+    """
+
+    def __init__(self, max_frames=256, **kwargs):
+        super().__init__(**kwargs)
+        self.max_frames = max_frames
+
+    async def _chunk(self, data: av.AudioFrame) -> Optional[list[av.AudioFrame]]:
+        self.frames.append(data)
+        if len(self.frames) >= self.max_frames:
+            frames_to_emit = self.frames[:]
+            self.frames = []
+            return frames_to_emit
+
+        return None
+
+    async def _flush(self):
+        frames = self.frames[:]
+        self.frames = []
+        if frames:
+            await self.emit(frames)
+
+
+AudioChunkerAutoProcessor.register("frames", AudioChunkerFramesProcessor)

server/reflector/processors/audio_chunker_silero.py

@@ -0,0 +1,293 @@
+from typing import Optional
+
+import av
+import numpy as np
+import torch
+from silero_vad import VADIterator, load_silero_vad
+
+from reflector.processors.audio_chunker import AudioChunkerProcessor
+from reflector.processors.audio_chunker_auto import AudioChunkerAutoProcessor
+
+
+class AudioChunkerSileroProcessor(AudioChunkerProcessor):
+    """
+    Assemble audio frames into chunks with VAD-based speech detection using Silero VAD.
+
+    Expects input audio to be already downscaled to 16kHz mono s16 format
+    (handled by AudioDownscaleProcessor in the pipeline).
+    """
+
+    def __init__(
+        self,
+        block_frames=256,
+        max_frames=1024,
+        use_onnx=True,
+        min_frames=2,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.block_frames = block_frames
+        self.max_frames = max_frames
+        self.min_frames = min_frames
+
+        # Initialize Silero VAD
+        self._init_vad(use_onnx)
+
+    def _init_vad(self, use_onnx=False):
+        """Initialize Silero VAD model for 16kHz audio"""
+        try:
+            torch.set_num_threads(1)
+            self.vad_model = load_silero_vad(onnx=use_onnx)
+            # VAD expects 16kHz audio (guaranteed by AudioDownscaleProcessor)
+            self.vad_iterator = VADIterator(self.vad_model, sampling_rate=16000)
+            self.logger.info("Silero VAD initialized for 16kHz audio")
+
+        except Exception as e:
+            self.logger.error(f"Failed to initialize Silero VAD: {e}")
+            self.vad_model = None
+            self.vad_iterator = None
+
+    async def _chunk(self, data: av.AudioFrame) -> Optional[list[av.AudioFrame]]:
+        """Process audio frame and return chunk when ready"""
+        self.frames.append(data)
+
+        # Check for speech segments every 32 frames (~1 second)
+        if len(self.frames) >= 32 and len(self.frames) % 32 == 0:
+            return await self._process_block()
+
+        # Safety fallback - emit if we hit max frames
+        elif len(self.frames) >= self.max_frames:
+            self.logger.warning(
+                f"AudioChunkerSileroProcessor: Reached max frames ({self.max_frames}), "
+                f"emitting first {self.max_frames // 2} frames"
+            )
+            frames_to_emit = self.frames[: self.max_frames // 2]
+            self.frames = self.frames[self.max_frames // 2 :]
+            if len(frames_to_emit) >= self.min_frames:
+                return frames_to_emit
+            else:
+                self.logger.debug(
+                    f"Ignoring fallback segment with {len(frames_to_emit)} frames "
+                    f"(< {self.min_frames} minimum)"
+                )
+
+        return None
+
+    async def _process_block(self) -> Optional[list[av.AudioFrame]]:
+        # Need at least 32 frames for VAD detection (~1 second)
+        if len(self.frames) < 32 or self.vad_iterator is None:
+            return None
+
+        # Processing block with current buffer size
+        # print(f"Processing block: {len(self.frames)} frames in buffer")
+
+        try:
+            # Convert frames to numpy array for VAD
+            audio_array = self._frames_to_numpy(self.frames)
+
+            if audio_array is None:
+                # Fallback: emit all frames if conversion failed
+                frames_to_emit = self.frames[:]
+                self.frames = []
+                if len(frames_to_emit) >= self.min_frames:
+                    return frames_to_emit
+                else:
+                    self.logger.debug(
+                        f"Ignoring conversion-failed segment with {len(frames_to_emit)} frames "
+                        f"(< {self.min_frames} minimum)"
+                    )
+                return None
+
+            # Find complete speech segments in the buffer
+            speech_end_frame = self._find_speech_segment_end(audio_array)
+
+            if speech_end_frame is None or speech_end_frame <= 0:
+                # No speech found but buffer is getting large
+                if len(self.frames) > 512:
+                    # Check if it's all silence and can be discarded
+                    # No speech segment found, buffer at {len(self.frames)} frames
+
+                    # Could emit silence or discard old frames here
+                    # For now, keep first 256 frames and discard older silence
+                    if len(self.frames) > 768:
+                        self.logger.debug(
+                            f"Discarding {len(self.frames) - 256} old frames (likely silence)"
+                        )
+                        self.frames = self.frames[-256:]
+                return None
+
+            # Calculate segment timing information
+            frames_to_emit = self.frames[:speech_end_frame]
+
+            # Get timing from av.AudioFrame
+            if frames_to_emit:
+                first_frame = frames_to_emit[0]
+                last_frame = frames_to_emit[-1]
+                sample_rate = first_frame.sample_rate
+
+                # Calculate duration
+                total_samples = sum(f.samples for f in frames_to_emit)
+                duration_seconds = total_samples / sample_rate if sample_rate > 0 else 0
+
+                # Get timestamps if available
+                start_time = (
+                    first_frame.pts * first_frame.time_base if first_frame.pts else 0
+                )
+                end_time = (
+                    last_frame.pts * last_frame.time_base if last_frame.pts else 0
+                )
+
+                # Convert to HH:MM:SS format for logging
+                def format_time(seconds):
+                    if not seconds:
+                        return "00:00:00"
+                    total_seconds = int(float(seconds))
+                    hours = total_seconds // 3600
+                    minutes = (total_seconds % 3600) // 60
+                    secs = total_seconds % 60
+                    return f"{hours:02d}:{minutes:02d}:{secs:02d}"
+
+                start_formatted = format_time(start_time)
+                end_formatted = format_time(end_time)
+
+                # Keep remaining frames for next processing
+                remaining_after = len(self.frames) - speech_end_frame
+
+                # Single structured log line
+                self.logger.info(
+                    "Speech segment found",
+                    start=start_formatted,
+                    end=end_formatted,
+                    frames=speech_end_frame,
+                    duration=round(duration_seconds, 2),
+                    buffer_before=len(self.frames),
+                    remaining=remaining_after,
+                )
+
+            # Keep remaining frames for next processing
+            self.frames = self.frames[speech_end_frame:]
+
+            # Filter out segments with too few frames
+            if len(frames_to_emit) >= self.min_frames:
+                return frames_to_emit
+            else:
+                self.logger.debug(
+                    f"Ignoring segment with {len(frames_to_emit)} frames "
+                    f"(< {self.min_frames} minimum)"
+                )
+
+        except Exception as e:
+            self.logger.error(f"Error in VAD processing: {e}")
+            # Fallback to simple chunking
+            if len(self.frames) >= self.block_frames:
+                frames_to_emit = self.frames[: self.block_frames]
+                self.frames = self.frames[self.block_frames :]
+                if len(frames_to_emit) >= self.min_frames:
+                    return frames_to_emit
+                else:
+                    self.logger.debug(
+                        f"Ignoring exception-fallback segment with {len(frames_to_emit)} frames "
+                        f"(< {self.min_frames} minimum)"
+                    )
+
+        return None
+
+    def _frames_to_numpy(self, frames: list[av.AudioFrame]) -> Optional[np.ndarray]:
+        """Convert av.AudioFrame list to numpy array for VAD processing
+
+        Input frames are already 16kHz mono s16 format from AudioDownscaleProcessor.
+        Only need to convert s16 to float32 for Silero VAD.
+        """
+        if not frames:
+            return None
+
+        try:
+            # Concatenate all frame arrays
+            audio_arrays = [frame.to_ndarray().flatten() for frame in frames]
+            if not audio_arrays:
+                return None
+
+            combined_audio = np.concatenate(audio_arrays)
+
+            # Convert s16 to float32 (Silero VAD requires float32 in range [-1.0, 1.0])
+            # Input is guaranteed to be s16 from AudioDownscaleProcessor
+            return combined_audio.astype(np.float32) / 32768.0
+
+        except Exception as e:
+            self.logger.error(f"Error converting frames to numpy: {e}")
+            return None
+
+    def _find_speech_segment_end(self, audio_array: np.ndarray) -> Optional[int]:
+        """Find complete speech segments and return frame index at segment end"""
+        if self.vad_iterator is None or len(audio_array) == 0:
+            return None
+
+        try:
+            # Process audio in 512-sample windows for VAD
+            window_size = 512
+            min_silence_windows = 3  # Require 3 windows of silence after speech
+
+            # Track speech state
+            in_speech = False
+            speech_start = None
+            speech_end = None
+            silence_count = 0
+
+            for i in range(0, len(audio_array), window_size):
+                chunk = audio_array[i : i + window_size]
+                if len(chunk) < window_size:
+                    chunk = np.pad(chunk, (0, window_size - len(chunk)))
+
+                # Detect if this window has speech
+                speech_dict = self.vad_iterator(chunk, return_seconds=True)
+
+                # VADIterator returns dict with 'start' and 'end' when speech segments are detected
+                if speech_dict:
+                    if not in_speech:
+                        # Speech started
+                        speech_start = i
+                        in_speech = True
+                        # Debug: print(f"Speech START at sample {i}, VAD: {speech_dict}")
+                    silence_count = 0  # Reset silence counter
+                    continue
+
+                if not in_speech:
+                    continue
+
+                # We're in speech but found silence
+                silence_count += 1
+                if silence_count < min_silence_windows:
+                    continue
+
+                # Found end of speech segment
+                speech_end = i - (min_silence_windows - 1) * window_size
+                # Debug: print(f"Speech END at sample {speech_end}")
+
+                # Convert sample position to frame index
+                samples_per_frame = self.frames[0].samples if self.frames else 1024
+                frame_index = speech_end // samples_per_frame
+
+                # Ensure we don't exceed buffer
+                frame_index = min(frame_index, len(self.frames))
+                return frame_index
+
+            return None
+
+        except Exception as e:
+            self.logger.error(f"Error finding speech segment: {e}")
+            return None
+
+    async def _flush(self):
+        frames = self.frames[:]
+        self.frames = []
+        if frames:
+            if len(frames) >= self.min_frames:
+                await self.emit(frames)
+            else:
+                self.logger.debug(
+                    f"Ignoring flush segment with {len(frames)} frames "
+                    f"(< {self.min_frames} minimum)"
+                )
+
+
+AudioChunkerAutoProcessor.register("silero", AudioChunkerSileroProcessor)

server/reflector/processors/audio_diarization.py

@@ -1,5 +1,10 @@
 from reflector.processors.base import Processor
-from reflector.processors.types import AudioDiarizationInput, TitleSummary, Word
+from reflector.processors.types import (
+    AudioDiarizationInput,
+    DiarizationSegment,
+    TitleSummary,
+    Word,
+)
 
 
 class AudioDiarizationProcessor(Processor):
@@ -33,18 +38,21 @@ class AudioDiarizationProcessor(Processor):
     async def _diarize(self, data: AudioDiarizationInput):
         raise NotImplementedError
 
-    def assign_speaker(self, words: list[Word], diarization: list[dict]):
-        self._diarization_remove_overlap(diarization)
-        self._diarization_remove_segment_without_words(words, diarization)
-        self._diarization_merge_same_speaker(words, diarization)
-        self._diarization_assign_speaker(words, diarization)
+    @classmethod
+    def assign_speaker(cls, words: list[Word], diarization: list[DiarizationSegment]):
+        cls._diarization_remove_overlap(diarization)
+        cls._diarization_remove_segment_without_words(words, diarization)
+        cls._diarization_merge_same_speaker(diarization)
+        cls._diarization_assign_speaker(words, diarization)
 
-    def iter_words_from_topics(self, topics: TitleSummary):
+    @staticmethod
+    def iter_words_from_topics(topics: list[TitleSummary]):
         for topic in topics:
             for word in topic.transcript.words:
                 yield word
 
-    def is_word_continuation(self, word_prev, word):
+    @staticmethod
+    def is_word_continuation(word_prev, word):
         """
         Return True if the word is a continuation of the previous word
         by checking if the previous word is ending with a punctuation
@@ -57,7 +65,8 @@ class AudioDiarizationProcessor(Processor):
             return False
         return True
 
-    def _diarization_remove_overlap(self, diarization: list[dict]):
+    @staticmethod
+    def _diarization_remove_overlap(diarization: list[DiarizationSegment]):
         """
         Remove overlap in diarization results
 
@@ -82,8 +91,9 @@ class AudioDiarizationProcessor(Processor):
             else:
                 diarization_idx += 1
 
+    @staticmethod
     def _diarization_remove_segment_without_words(
-        self, words: list[Word], diarization: list[dict]
+        words: list[Word], diarization: list[DiarizationSegment]
     ):
         """
         Remove diarization segments without words
@@ -112,9 +122,8 @@ class AudioDiarizationProcessor(Processor):
             else:
                 diarization_idx += 1
 
-    def _diarization_merge_same_speaker(
-        self, words: list[Word], diarization: list[dict]
-    ):
+    @staticmethod
+    def _diarization_merge_same_speaker(diarization: list[DiarizationSegment]):
         """
         Merge diarization contigous segments with the same speaker
 
@@ -131,7 +140,10 @@ class AudioDiarizationProcessor(Processor):
             else:
                 diarization_idx += 1
 
-    def _diarization_assign_speaker(self, words: list[Word], diarization: list[dict]):
+    @classmethod
+    def _diarization_assign_speaker(
+        cls, words: list[Word], diarization: list[DiarizationSegment]
+    ):
         """
         Assign speaker to words based on diarization
 
@@ -139,7 +151,7 @@ class AudioDiarizationProcessor(Processor):
         """
 
         word_idx = 0
-        last_speaker = None
+        last_speaker = 0
         for d in diarization:
             start = d["start"]
             end = d["end"]
@@ -154,7 +166,7 @@ class AudioDiarizationProcessor(Processor):
                     # If it's a continuation, assign with the last speaker
                     is_continuation = False
                     if word_idx > 0 and word_idx < len(words) - 1:
-                        is_continuation = self.is_word_continuation(
+                        is_continuation = cls.is_word_continuation(
                             *words[word_idx - 1 : word_idx + 1]
                         )
                     if is_continuation:

server/reflector/processors/audio_diarization_pyannote.py

@@ -0,0 +1,74 @@
+import os
+
+import torch
+import torchaudio
+from pyannote.audio import Pipeline
+
+from reflector.processors.audio_diarization import AudioDiarizationProcessor
+from reflector.processors.audio_diarization_auto import AudioDiarizationAutoProcessor
+from reflector.processors.types import AudioDiarizationInput, DiarizationSegment
+
+
+class AudioDiarizationPyannoteProcessor(AudioDiarizationProcessor):
+    """Local diarization processor using pyannote.audio library"""
+
+    def __init__(
+        self,
+        model_name: str = "pyannote/speaker-diarization-3.1",
+        pyannote_auth_token: str | None = None,
+        device: str | None = None,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.model_name = model_name
+        self.auth_token = pyannote_auth_token or os.environ.get("HF_TOKEN")
+        self.device = device
+
+        if device is None:
+            self.device = "cuda" if torch.cuda.is_available() else "cpu"
+
+        self.logger.info(f"Loading pyannote diarization model: {self.model_name}")
+        self.diarization_pipeline = Pipeline.from_pretrained(
+            self.model_name, use_auth_token=self.auth_token
+        )
+        self.diarization_pipeline.to(torch.device(self.device))
+        self.logger.info(f"Diarization model loaded on device: {self.device}")
+
+    async def _diarize(self, data: AudioDiarizationInput) -> list[DiarizationSegment]:
+        try:
+            # Load audio file (audio_url is assumed to be a local file path)
+            self.logger.info(f"Loading local audio file: {data.audio_url}")
+            waveform, sample_rate = torchaudio.load(data.audio_url)
+            audio_input = {"waveform": waveform, "sample_rate": sample_rate}
+            self.logger.info("Running speaker diarization")
+            diarization = self.diarization_pipeline(audio_input)
+
+            # Convert pyannote diarization output to our format
+            segments = []
+            for segment, _, speaker in diarization.itertracks(yield_label=True):
+                # Extract speaker number from label (e.g., "SPEAKER_00" -> 0)
+                speaker_id = 0
+                if speaker.startswith("SPEAKER_"):
+                    try:
+                        speaker_id = int(speaker.split("_")[-1])
+                    except (ValueError, IndexError):
+                        # Fallback to hash-based ID if parsing fails
+                        speaker_id = hash(speaker) % 1000
+
+                segments.append(
+                    {
+                        "start": round(segment.start, 3),
+                        "end": round(segment.end, 3),
+                        "speaker": speaker_id,
+                    }
+                )
+
+            self.logger.info(f"Diarization completed with {len(segments)} segments")
+            return segments
+
+        except Exception as e:
+            self.logger.exception(f"Diarization failed: {e}")
+            raise
+
+
+AudioDiarizationAutoProcessor.register("pyannote", AudioDiarizationPyannoteProcessor)

server/reflector/processors/audio_downscale.py

@@ -0,0 +1,60 @@
+from typing import Optional
+
+import av
+from av.audio.resampler import AudioResampler
+
+from reflector.processors.base import Processor
+
+
+def copy_frame(frame: av.AudioFrame) -> av.AudioFrame:
+    frame_copy = frame.from_ndarray(
+        frame.to_ndarray(),
+        format=frame.format.name,
+        layout=frame.layout.name,
+    )
+    frame_copy.sample_rate = frame.sample_rate
+    frame_copy.pts = frame.pts
+    frame_copy.time_base = frame.time_base
+    return frame_copy
+
+
+class AudioDownscaleProcessor(Processor):
+    """
+    Downscale audio frames to 16kHz mono format
+    """
+
+    INPUT_TYPE = av.AudioFrame
+    OUTPUT_TYPE = av.AudioFrame
+
+    def __init__(self, target_rate: int = 16000, target_layout: str = "mono", **kwargs):
+        super().__init__(**kwargs)
+        self.target_rate = target_rate
+        self.target_layout = target_layout
+        self.resampler: Optional[AudioResampler] = None
+        self.needs_resampling: Optional[bool] = None
+
+    async def _push(self, data: av.AudioFrame):
+        if self.needs_resampling is None:
+            self.needs_resampling = (
+                data.sample_rate != self.target_rate
+                or data.layout.name != self.target_layout
+            )
+
+            if self.needs_resampling:
+                self.resampler = AudioResampler(
+                    format="s16", layout=self.target_layout, rate=self.target_rate
+                )
+
+        if not self.needs_resampling or not self.resampler:
+            await self.emit(data)
+            return
+
+        resampled_frames = self.resampler.resample(copy_frame(data))
+        for resampled_frame in resampled_frames:
+            await self.emit(resampled_frame)
+
+    async def _flush(self):
+        if self.needs_resampling and self.resampler:
+            final_frames = self.resampler.resample(None)
+            for frame in final_frames:
+                await self.emit(frame)

server/reflector/processors/audio_merge.py

@@ -16,37 +16,46 @@ class AudioMergeProcessor(Processor):
     INPUT_TYPE = list[av.AudioFrame]
     OUTPUT_TYPE = AudioFile
 
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
     async def _push(self, data: list[av.AudioFrame]):
         if not data:
             return
 
         # get audio information from first frame
         frame = data[0]
-        channels = len(frame.layout.channels)
-        sample_rate = frame.sample_rate
-        sample_width = frame.format.bytes
+        output_channels = len(frame.layout.channels)
+        output_sample_rate = frame.sample_rate
+        output_sample_width = frame.format.bytes
 
         # create audio file
         uu = uuid4().hex
         fd = io.BytesIO()
 
+        # Use PyAV to write frames
         out_container = av.open(fd, "w", format="wav")
-        out_stream = out_container.add_stream("pcm_s16le", rate=sample_rate)
+        out_stream = out_container.add_stream("pcm_s16le", rate=output_sample_rate)
+        out_stream.layout = frame.layout.name
+
         for frame in data:
             for packet in out_stream.encode(frame):
                 out_container.mux(packet)
+
+        # Flush the encoder
         for packet in out_stream.encode(None):
             out_container.mux(packet)
         out_container.close()
+
         fd.seek(0)
 
         # emit audio file
         audiofile = AudioFile(
             name=f"{monotonic_ns()}-{uu}.wav",
             fd=fd,
-            sample_rate=sample_rate,
-            channels=channels,
-            sample_width=sample_width,
+            sample_rate=output_sample_rate,
+            channels=output_channels,
+            sample_width=output_sample_width,
             timestamp=data[0].pts * data[0].time_base,
         )
 

server/reflector/processors/audio_transcript_modal.py

@@ -21,7 +21,11 @@ from reflector.settings import settings
 
 
 class AudioTranscriptModalProcessor(AudioTranscriptProcessor):
-    def __init__(self, modal_api_key: str | None = None, **kwargs):
+    def __init__(
+        self,
+        modal_api_key: str | None = None,
+        **kwargs,
+    ):
         super().__init__()
         if not settings.TRANSCRIPT_URL:
             raise Exception(

server/reflector/processors/base.py

@@ -173,6 +173,7 @@ class Processor(Emitter):
         except Exception:
             self.m_processor_failure.inc()
             self.logger.exception("Error in push")
+            raise
 
     async def flush(self):
         """
@@ -240,14 +241,15 @@ class ThreadedProcessor(Processor):
         self.INPUT_TYPE = processor.INPUT_TYPE
         self.OUTPUT_TYPE = processor.OUTPUT_TYPE
         self.executor = ThreadPoolExecutor(max_workers=max_workers)
-        self.queue = asyncio.Queue()
-        self.task = asyncio.get_running_loop().create_task(self.loop())
+        self.queue = asyncio.Queue(maxsize=50)
+        self.task: asyncio.Task | None = None
 
     def set_pipeline(self, pipeline: "Pipeline"):
         super().set_pipeline(pipeline)
         self.processor.set_pipeline(pipeline)
 
     async def loop(self):
+        try:
             while True:
                 data = await self.queue.get()
                 self.m_processor_queue.set(self.queue.qsize())
@@ -265,8 +267,19 @@ class ThreadedProcessor(Processor):
                             )
                     finally:
                         self.queue.task_done()
+        except Exception as e:
+            logger.error(f"Crash in {self.__class__.__name__}: {e}", exc_info=e)
+
+    async def _ensure_task(self):
+        if self.task is None:
+            self.task = asyncio.get_running_loop().create_task(self.loop())
+
+        # XXX not doing a sleep here make the whole pipeline prior the thread
+        # to be running without having a chance to work on the task here.
+        await asyncio.sleep(0)
 
     async def _push(self, data):
+        await self._ensure_task()
         await self.queue.put(data)
 
     async def _flush(self):

server/reflector/processors/file_diarization.py

@@ -0,0 +1,33 @@
+from pydantic import BaseModel
+
+from reflector.processors.base import Processor
+from reflector.processors.types import DiarizationSegment
+
+
+class FileDiarizationInput(BaseModel):
+    """Input for file diarization containing audio URL"""
+
+    audio_url: str
+
+
+class FileDiarizationOutput(BaseModel):
+    """Output for file diarization containing speaker segments"""
+
+    diarization: list[DiarizationSegment]
+
+
+class FileDiarizationProcessor(Processor):
+    """
+    Diarize complete audio files from URL
+    """
+
+    INPUT_TYPE = FileDiarizationInput
+    OUTPUT_TYPE = FileDiarizationOutput
+
+    async def _push(self, data: FileDiarizationInput):
+        result = await self._diarize(data)
+        if result:
+            await self.emit(result)
+
+    async def _diarize(self, data: FileDiarizationInput):
+        raise NotImplementedError

server/reflector/processors/file_diarization_auto.py

@@ -0,0 +1,33 @@
+import importlib
+
+from reflector.processors.file_diarization import FileDiarizationProcessor
+from reflector.settings import settings
+
+
+class FileDiarizationAutoProcessor(FileDiarizationProcessor):
+    _registry = {}
+
+    @classmethod
+    def register(cls, name, kclass):
+        cls._registry[name] = kclass
+
+    def __new__(cls, name: str | None = None, **kwargs):
+        if name is None:
+            name = settings.DIARIZATION_BACKEND
+
+        if name not in cls._registry:
+            module_name = f"reflector.processors.file_diarization_{name}"
+            importlib.import_module(module_name)
+
+        # gather specific configuration for the processor
+        # search `DIARIZATION_BACKEND_XXX_YYY`, push to constructor as `backend_xxx_yyy`
+        config = {}
+        name_upper = name.upper()
+        settings_prefix = "DIARIZATION_"
+        config_prefix = f"{settings_prefix}{name_upper}_"
+        for key, value in settings:
+            if key.startswith(config_prefix):
+                config_name = key[len(settings_prefix) :].lower()
+                config[config_name] = value
+
+        return cls._registry[name](**config | kwargs)

server/reflector/processors/file_diarization_modal.py

@@ -0,0 +1,57 @@
+"""
+File diarization implementation using the GPU service from modal.com
+
+API will be a POST request to DIARIZATION_URL:
+
+```
+POST /diarize?audio_file_url=...&timestamp=0
+Authorization: Bearer <modal_api_key>
+```
+"""
+
+import httpx
+
+from reflector.processors.file_diarization import (
+    FileDiarizationInput,
+    FileDiarizationOutput,
+    FileDiarizationProcessor,
+)
+from reflector.processors.file_diarization_auto import FileDiarizationAutoProcessor
+from reflector.settings import settings
+
+
+class FileDiarizationModalProcessor(FileDiarizationProcessor):
+    def __init__(self, modal_api_key: str | None = None, **kwargs):
+        super().__init__(**kwargs)
+        if not settings.DIARIZATION_URL:
+            raise Exception(
+                "DIARIZATION_URL required to use FileDiarizationModalProcessor"
+            )
+        self.diarization_url = settings.DIARIZATION_URL + "/diarize"
+        self.file_timeout = settings.DIARIZATION_FILE_TIMEOUT
+        self.modal_api_key = modal_api_key
+
+    async def _diarize(self, data: FileDiarizationInput):
+        """Get speaker diarization for file"""
+        self.logger.info(f"Starting diarization from {data.audio_url}")
+
+        headers = {}
+        if self.modal_api_key:
+            headers["Authorization"] = f"Bearer {self.modal_api_key}"
+
+        async with httpx.AsyncClient(timeout=self.file_timeout) as client:
+            response = await client.post(
+                self.diarization_url,
+                headers=headers,
+                params={
+                    "audio_file_url": data.audio_url,
+                    "timestamp": 0,
+                },
+            )
+            response.raise_for_status()
+            diarization_data = response.json()["diarization"]
+
+        return FileDiarizationOutput(diarization=diarization_data)
+
+
+FileDiarizationAutoProcessor.register("modal", FileDiarizationModalProcessor)

server/reflector/processors/file_transcript.py

@@ -0,0 +1,65 @@
+from prometheus_client import Counter, Histogram
+
+from reflector.processors.base import Processor
+from reflector.processors.types import Transcript
+
+
+class FileTranscriptInput:
+    """Input for file transcription containing audio URL and language settings"""
+
+    def __init__(self, audio_url: str, language: str = "en"):
+        self.audio_url = audio_url
+        self.language = language
+
+
+class FileTranscriptProcessor(Processor):
+    """
+    Transcript complete audio files from URL
+    """
+
+    INPUT_TYPE = FileTranscriptInput
+    OUTPUT_TYPE = Transcript
+
+    m_transcript = Histogram(
+        "file_transcript",
+        "Time spent in FileTranscript.transcript",
+        ["backend"],
+    )
+    m_transcript_call = Counter(
+        "file_transcript_call",
+        "Number of calls to FileTranscript.transcript",
+        ["backend"],
+    )
+    m_transcript_success = Counter(
+        "file_transcript_success",
+        "Number of successful calls to FileTranscript.transcript",
+        ["backend"],
+    )
+    m_transcript_failure = Counter(
+        "file_transcript_failure",
+        "Number of failed calls to FileTranscript.transcript",
+        ["backend"],
+    )
+
+    def __init__(self, *args, **kwargs):
+        name = self.__class__.__name__
+        self.m_transcript = self.m_transcript.labels(name)
+        self.m_transcript_call = self.m_transcript_call.labels(name)
+        self.m_transcript_success = self.m_transcript_success.labels(name)
+        self.m_transcript_failure = self.m_transcript_failure.labels(name)
+        super().__init__(*args, **kwargs)
+
+    async def _push(self, data: FileTranscriptInput):
+        try:
+            self.m_transcript_call.inc()
+            with self.m_transcript.time():
+                result = await self._transcript(data)
+            self.m_transcript_success.inc()
+            if result:
+                await self.emit(result)
+        except Exception:
+            self.m_transcript_failure.inc()
+            raise
+
+    async def _transcript(self, data: FileTranscriptInput):
+        raise NotImplementedError

server/reflector/processors/file_transcript_auto.py

@@ -0,0 +1,32 @@
+import importlib
+
+from reflector.processors.file_transcript import FileTranscriptProcessor
+from reflector.settings import settings
+
+
+class FileTranscriptAutoProcessor(FileTranscriptProcessor):
+    _registry = {}
+
+    @classmethod
+    def register(cls, name, kclass):
+        cls._registry[name] = kclass
+
+    def __new__(cls, name: str | None = None, **kwargs):
+        if name is None:
+            name = settings.TRANSCRIPT_BACKEND
+        if name not in cls._registry:
+            module_name = f"reflector.processors.file_transcript_{name}"
+            importlib.import_module(module_name)
+
+        # gather specific configuration for the processor
+        # search `TRANSCRIPT_BACKEND_XXX_YYY`, push to constructor as `backend_xxx_yyy`
+        config = {}
+        name_upper = name.upper()
+        settings_prefix = "TRANSCRIPT_"
+        config_prefix = f"{settings_prefix}{name_upper}_"
+        for key, value in settings:
+            if key.startswith(config_prefix):
+                config_name = key[len(settings_prefix) :].lower()
+                config[config_name] = value
+
+        return cls._registry[name](**config | kwargs)

server/reflector/processors/file_transcript_modal.py

@@ -0,0 +1,74 @@
+"""
+File transcription implementation using the GPU service from modal.com
+
+API will be a POST request to TRANSCRIPT_URL:
+
+```json
+{
+    "audio_file_url": "https://...",
+    "language": "en",
+    "model": "parakeet-tdt-0.6b-v2",
+    "batch": true
+}
+```
+"""
+
+import httpx
+
+from reflector.processors.file_transcript import (
+    FileTranscriptInput,
+    FileTranscriptProcessor,
+)
+from reflector.processors.file_transcript_auto import FileTranscriptAutoProcessor
+from reflector.processors.types import Transcript, Word
+from reflector.settings import settings
+
+
+class FileTranscriptModalProcessor(FileTranscriptProcessor):
+    def __init__(self, modal_api_key: str | None = None, **kwargs):
+        super().__init__(**kwargs)
+        if not settings.TRANSCRIPT_URL:
+            raise Exception(
+                "TRANSCRIPT_URL required to use FileTranscriptModalProcessor"
+            )
+        self.transcript_url = settings.TRANSCRIPT_URL
+        self.file_timeout = settings.TRANSCRIPT_FILE_TIMEOUT
+        self.modal_api_key = modal_api_key
+
+    async def _transcript(self, data: FileTranscriptInput):
+        """Send full file to Modal for transcription"""
+        url = f"{self.transcript_url}/v1/audio/transcriptions-from-url"
+
+        self.logger.info(f"Starting file transcription from {data.audio_url}")
+
+        headers = {}
+        if self.modal_api_key:
+            headers["Authorization"] = f"Bearer {self.modal_api_key}"
+
+        async with httpx.AsyncClient(timeout=self.file_timeout) as client:
+            response = await client.post(
+                url,
+                headers=headers,
+                json={
+                    "audio_file_url": data.audio_url,
+                    "language": data.language,
+                    "batch": True,
+                },
+            )
+            response.raise_for_status()
+            result = response.json()
+
+        words = [
+            Word(
+                text=word_info["word"],
+                start=word_info["start"],
+                end=word_info["end"],
+            )
+            for word_info in result.get("words", [])
+        ]
+
+        return Transcript(words=words)
+
+
+# Register with the auto processor
+FileTranscriptAutoProcessor.register("modal", FileTranscriptModalProcessor)

server/reflector/processors/summary/summary_builder.py

@@ -6,7 +6,7 @@ This script is used to generate a summary of a meeting notes transcript.
 
 import asyncio
 import sys
-from datetime import datetime
+from datetime import datetime, timezone
 from enum import Enum
 from textwrap import dedent
 from typing import Type, TypeVar
@@ -474,7 +474,7 @@ if __name__ == "__main__":
 
         if args.save:
             # write the summary to a file, on the format summary-<iso date>.md
-            filename = f"summary-{datetime.now().isoformat()}.md"
+            filename = f"summary-{datetime.now(timezone.utc).isoformat()}.md"
             with open(filename, "w", encoding="utf-8") as f:
                 f.write(sm.as_markdown())
 

server/reflector/processors/transcript_diarization_assembler.py

@@ -0,0 +1,45 @@
+"""
+Processor to assemble transcript with diarization results
+"""
+
+from reflector.processors.audio_diarization import AudioDiarizationProcessor
+from reflector.processors.base import Processor
+from reflector.processors.types import DiarizationSegment, Transcript
+
+
+class TranscriptDiarizationAssemblerInput:
+    """Input containing transcript and diarization data"""
+
+    def __init__(self, transcript: Transcript, diarization: list[DiarizationSegment]):
+        self.transcript = transcript
+        self.diarization = diarization
+
+
+class TranscriptDiarizationAssemblerProcessor(Processor):
+    """
+    Assemble transcript with diarization results by applying speaker assignments
+    """
+
+    INPUT_TYPE = TranscriptDiarizationAssemblerInput
+    OUTPUT_TYPE = Transcript
+
+    async def _push(self, data: TranscriptDiarizationAssemblerInput):
+        result = await self._assemble(data)
+        if result:
+            await self.emit(result)
+
+    async def _assemble(self, data: TranscriptDiarizationAssemblerInput):
+        """Apply diarization to transcript words"""
+        if not data.diarization:
+            self.logger.info(
+                "No diarization data provided, returning original transcript"
+            )
+            return data.transcript
+
+        # Reuse logic from AudioDiarizationProcessor
+        processor = AudioDiarizationProcessor()
+        words = data.transcript.words
+        processor.assign_speaker(words, data.diarization)
+
+        self.logger.info(f"Applied diarization to {len(words)} words")
+        return data.transcript

server/reflector/processors/types.py

@@ -2,12 +2,22 @@ import io
 import re
 import tempfile
 from pathlib import Path
+from typing import Annotated, TypedDict
 
 from profanityfilter import ProfanityFilter
-from pydantic import BaseModel, PrivateAttr
+from pydantic import BaseModel, Field, PrivateAttr
 
 from reflector.redis_cache import redis_cache
 
+
+class DiarizationSegment(TypedDict):
+    """Type definition for diarization segment containing speaker information"""
+
+    start: float
+    end: float
+    speaker: int
+
+
 PUNC_RE = re.compile(r"[.;:?!…]")
 
 profanity_filter = ProfanityFilter()
@@ -48,20 +58,70 @@ class AudioFile(BaseModel):
             self._path.unlink()
 
 
+# non-negative seconds with float part
+Seconds = Annotated[float, Field(ge=0.0, description="Time in seconds with float part")]
+
+
 class Word(BaseModel):
     text: str
-    start: float
-    end: float
+    start: Seconds
+    end: Seconds
     speaker: int = 0
 
 
 class TranscriptSegment(BaseModel):
     text: str
-    start: float
-    end: float
+    start: Seconds
+    end: Seconds
     speaker: int = 0
 
 
+def words_to_segments(words: list[Word]) -> list[TranscriptSegment]:
+    # from a list of word, create a list of segments
+    # join the word that are less than 2 seconds apart
+    # but separate if the speaker changes, or if the punctuation is a . , ; : ? !
+    segments = []
+    current_segment = None
+    MAX_SEGMENT_LENGTH = 120
+
+    for word in words:
+        if current_segment is None:
+            current_segment = TranscriptSegment(
+                text=word.text,
+                start=word.start,
+                end=word.end,
+                speaker=word.speaker,
+            )
+            continue
+
+        # If the word is attach to another speaker, push the current segment
+        # and start a new one
+        if word.speaker != current_segment.speaker:
+            segments.append(current_segment)
+            current_segment = TranscriptSegment(
+                text=word.text,
+                start=word.start,
+                end=word.end,
+                speaker=word.speaker,
+            )
+            continue
+
+        # if the word is the end of a sentence, and we have enough content,
+        # add the word to the current segment and push it
+        current_segment.text += word.text
+        current_segment.end = word.end
+
+        have_punc = PUNC_RE.search(word.text)
+        if have_punc and (len(current_segment.text) > MAX_SEGMENT_LENGTH):
+            segments.append(current_segment)
+            current_segment = None
+
+    if current_segment:
+        segments.append(current_segment)
+
+    return segments
+
+
 class Transcript(BaseModel):
     translation: str | None = None
     words: list[Word] = None
@@ -117,49 +177,7 @@ class Transcript(BaseModel):
         return Transcript(text=self.text, translation=self.translation, words=words)
 
     def as_segments(self) -> list[TranscriptSegment]:
-        # from a list of word, create a list of segments
-        # join the word that are less than 2 seconds apart
-        # but separate if the speaker changes, or if the punctuation is a . , ; : ? !
-        segments = []
-        current_segment = None
-        MAX_SEGMENT_LENGTH = 120
-
-        for word in self.words:
-            if current_segment is None:
-                current_segment = TranscriptSegment(
-                    text=word.text,
-                    start=word.start,
-                    end=word.end,
-                    speaker=word.speaker,
-                )
-                continue
-
-            # If the word is attach to another speaker, push the current segment
-            # and start a new one
-            if word.speaker != current_segment.speaker:
-                segments.append(current_segment)
-                current_segment = TranscriptSegment(
-                    text=word.text,
-                    start=word.start,
-                    end=word.end,
-                    speaker=word.speaker,
-                )
-                continue
-
-            # if the word is the end of a sentence, and we have enough content,
-            # add the word to the current segment and push it
-            current_segment.text += word.text
-            current_segment.end = word.end
-
-            have_punc = PUNC_RE.search(word.text)
-            if have_punc and (len(current_segment.text) > MAX_SEGMENT_LENGTH):
-                segments.append(current_segment)
-                current_segment = None
-
-        if current_segment:
-            segments.append(current_segment)
-
-        return segments
+        return words_to_segments(self.words)
 
 
 class TitleSummary(BaseModel):

server/reflector/settings.py

@@ -14,16 +14,23 @@ class Settings(BaseSettings):
     CORS_ALLOW_CREDENTIALS: bool = False
 
     # Database
-    DATABASE_URL: str = "sqlite:///./reflector.sqlite3"
+    DATABASE_URL: str = (
+        "postgresql+asyncpg://reflector:reflector@localhost:5432/reflector"
+    )
 
     # local data directory
     DATA_DIR: str = "./data"
 
+    # Audio Chunking
+    # backends: silero, frames
+    AUDIO_CHUNKER_BACKEND: str = "frames"
+
     # Audio Transcription
     # backends: whisper, modal
     TRANSCRIPT_BACKEND: str = "whisper"
     TRANSCRIPT_URL: str | None = None
     TRANSCRIPT_TIMEOUT: int = 90
+    TRANSCRIPT_FILE_TIMEOUT: int = 600
 
     # Audio Transcription: modal backend
     TRANSCRIPT_MODAL_API_KEY: str | None = None
@@ -37,6 +44,15 @@ class Settings(BaseSettings):
     TRANSCRIPT_STORAGE_AWS_ACCESS_KEY_ID: str | None = None
     TRANSCRIPT_STORAGE_AWS_SECRET_ACCESS_KEY: str | None = None
 
+    # Recording storage
+    RECORDING_STORAGE_BACKEND: str | None = None
+
+    # Recording storage configuration for AWS
+    RECORDING_STORAGE_AWS_BUCKET_NAME: str = "recording-bucket"
+    RECORDING_STORAGE_AWS_REGION: str = "us-east-1"
+    RECORDING_STORAGE_AWS_ACCESS_KEY_ID: str | None = None
+    RECORDING_STORAGE_AWS_SECRET_ACCESS_KEY: str | None = None
+
     # Translate into the target language
     TRANSLATION_BACKEND: str = "passthrough"
     TRANSLATE_URL: str | None = None
@@ -55,10 +71,14 @@ class Settings(BaseSettings):
     DIARIZATION_ENABLED: bool = True
     DIARIZATION_BACKEND: str = "modal"
     DIARIZATION_URL: str | None = None
+    DIARIZATION_FILE_TIMEOUT: int = 600
 
     # Diarization: modal backend
     DIARIZATION_MODAL_API_KEY: str | None = None
 
+    # Diarization: local pyannote.audio
+    DIARIZATION_PYANNOTE_AUTH_TOKEN: str | None = None
+
     # Sentry
     SENTRY_DSN: str | None = None
 
@@ -102,7 +122,6 @@ class Settings(BaseSettings):
     WHEREBY_API_URL: str = "https://api.whereby.dev/v1"
     WHEREBY_API_KEY: str | None = None
     WHEREBY_WEBHOOK_SECRET: str | None = None
-    AWS_WHEREBY_S3_BUCKET: str | None = None
     AWS_WHEREBY_ACCESS_KEY_ID: str | None = None
     AWS_WHEREBY_ACCESS_KEY_SECRET: str | None = None
     AWS_PROCESS_RECORDING_QUEUE_URL: str | None = None

server/reflector/storage/__init__.py

@@ -1,10 +1,17 @@
 from .base import Storage  # noqa
+from reflector.settings import settings
 
 
 def get_transcripts_storage() -> Storage:
-    from reflector.settings import settings
-
+    assert settings.TRANSCRIPT_STORAGE_BACKEND
     return Storage.get_instance(
         name=settings.TRANSCRIPT_STORAGE_BACKEND,
         settings_prefix="TRANSCRIPT_STORAGE_",
     )
+
+
+def get_recordings_storage() -> Storage:
+    return Storage.get_instance(
+        name=settings.RECORDING_STORAGE_BACKEND,
+        settings_prefix="RECORDING_STORAGE_",
+    )

server/reflector/tools/exportdanswer.py

@@ -9,8 +9,9 @@ async def export_db(filename: str) -> None:
     filename = pathlib.Path(filename).resolve()
     settings.DATABASE_URL = f"sqlite:///{filename}"
 
-    from reflector.db import database, transcripts
+    from reflector.db import get_database, transcripts
 
+    database = get_database()
     await database.connect()
     transcripts = await database.fetch_all(transcripts.select())
     await database.disconnect()

server/reflector/tools/exportdb.py

@@ -8,8 +8,9 @@ async def export_db(filename: str) -> None:
     filename = pathlib.Path(filename).resolve()
     settings.DATABASE_URL = f"sqlite:///{filename}"
 
-    from reflector.db import database, transcripts
+    from reflector.db import get_database, transcripts
 
+    database = get_database()
     await database.connect()
     transcripts = await database.fetch_all(transcripts.select())
     await database.disconnect()

server/reflector/tools/process.py

@@ -1,10 +1,24 @@
+"""
+Process audio file with diarization support
+===========================================
+
+Extended version of process.py that includes speaker diarization.
+This tool processes audio files locally without requiring the full server infrastructure.
+"""
+
 import asyncio
+import tempfile
+import uuid
+from pathlib import Path
+from typing import List
 
 import av
 
 from reflector.logger import logger
 from reflector.processors import (
-    AudioChunkerProcessor,
+    AudioChunkerAutoProcessor,
+    AudioDownscaleProcessor,
+    AudioFileWriterProcessor,
     AudioMergeProcessor,
     AudioTranscriptAutoProcessor,
     Pipeline,
@@ -15,7 +29,43 @@ from reflector.processors import (
     TranscriptTopicDetectorProcessor,
     TranscriptTranslatorAutoProcessor,
 )
-from reflector.processors.base import BroadcastProcessor
+from reflector.processors.base import BroadcastProcessor, Processor
+from reflector.processors.types import (
+    AudioDiarizationInput,
+    TitleSummary,
+    TitleSummaryWithId,
+)
+
+
+class TopicCollectorProcessor(Processor):
+    """Collect topics for diarization"""
+
+    INPUT_TYPE = TitleSummary
+    OUTPUT_TYPE = TitleSummary
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.topics: List[TitleSummaryWithId] = []
+        self._topic_id = 0
+
+    async def _push(self, data: TitleSummary):
+        # Convert to TitleSummaryWithId and collect
+        self._topic_id += 1
+        topic_with_id = TitleSummaryWithId(
+            id=str(self._topic_id),
+            title=data.title,
+            summary=data.summary,
+            timestamp=data.timestamp,
+            duration=data.duration,
+            transcript=data.transcript,
+        )
+        self.topics.append(topic_with_id)
+
+        # Pass through the original topic
+        await self.emit(data)
+
+    def get_topics(self) -> List[TitleSummaryWithId]:
+        return self.topics
 
 
 async def process_audio_file(
@@ -24,18 +74,41 @@ async def process_audio_file(
     only_transcript=False,
     source_language="en",
     target_language="en",
+    enable_diarization=True,
+    diarization_backend="pyannote",
 ):
-    # build pipeline for audio processing
-    processors = [
-        AudioChunkerProcessor(),
+    # Create temp file for audio if diarization is enabled
+    audio_temp_path = None
+    if enable_diarization:
+        audio_temp_file = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+        audio_temp_path = audio_temp_file.name
+        audio_temp_file.close()
+
+    # Create processor for collecting topics
+    topic_collector = TopicCollectorProcessor()
+
+    # Build pipeline for audio processing
+    processors = []
+
+    # Add audio file writer at the beginning if diarization is enabled
+    if enable_diarization:
+        processors.append(AudioFileWriterProcessor(audio_temp_path))
+
+    # Add the rest of the processors
+    processors += [
+        AudioDownscaleProcessor(),
+        AudioChunkerAutoProcessor(),
         AudioMergeProcessor(),
         AudioTranscriptAutoProcessor.as_threaded(),
         TranscriptLinerProcessor(),
         TranscriptTranslatorAutoProcessor.as_threaded(),
     ]
+
     if not only_transcript:
         processors += [
             TranscriptTopicDetectorProcessor.as_threaded(),
+            # Collect topics for diarization
+            topic_collector,
             BroadcastProcessor(
                 processors=[
                     TranscriptFinalTitleProcessor.as_threaded(),
@@ -44,14 +117,14 @@ async def process_audio_file(
             ),
         ]
 
-    # transcription output
+    # Create main pipeline
     pipeline = Pipeline(*processors)
     pipeline.set_pref("audio:source_language", source_language)
     pipeline.set_pref("audio:target_language", target_language)
     pipeline.describe()
     pipeline.on(event_callback)
 
-    # start processing audio
+    # Start processing audio
     logger.info(f"Opening {filename}")
     container = av.open(filename)
     try:
@@ -62,34 +135,220 @@ async def process_audio_file(
         logger.info("Flushing the pipeline")
         await pipeline.flush()
 
-    logger.info("All done !")
+    # Run diarization if enabled and we have topics
+    if enable_diarization and not only_transcript and audio_temp_path:
+        topics = topic_collector.get_topics()
+
+        if topics:
+            logger.info(f"Starting diarization with {len(topics)} topics")
+
+            try:
+                from reflector.processors import AudioDiarizationAutoProcessor
+
+                diarization_processor = AudioDiarizationAutoProcessor(
+                    name=diarization_backend
+                )
+
+                diarization_processor.set_pipeline(pipeline)
+
+                # For Modal backend, we need to upload the file to S3 first
+                if diarization_backend == "modal":
+                    from datetime import datetime
+
+                    from reflector.storage import get_transcripts_storage
+                    from reflector.utils.s3_temp_file import S3TemporaryFile
+
+                    storage = get_transcripts_storage()
+
+                    # Generate a unique filename in evaluation folder
+                    timestamp = datetime.utcnow().strftime("%Y%m%d_%H%M%S")
+                    audio_filename = f"evaluation/diarization_temp/{timestamp}_{uuid.uuid4().hex}.wav"
+
+                    # Use context manager for automatic cleanup
+                    async with S3TemporaryFile(storage, audio_filename) as s3_file:
+                        # Read and upload the audio file
+                        with open(audio_temp_path, "rb") as f:
+                            audio_data = f.read()
+
+                        audio_url = await s3_file.upload(audio_data)
+                        logger.info(f"Uploaded audio to S3: {audio_filename}")
+
+                        # Create diarization input with S3 URL
+                        diarization_input = AudioDiarizationInput(
+                            audio_url=audio_url, topics=topics
+                        )
+
+                        # Run diarization
+                        await diarization_processor.push(diarization_input)
+                        await diarization_processor.flush()
+
+                        logger.info("Diarization complete")
+                        # File will be automatically cleaned up when exiting the context
+                else:
+                    # For local backend, use local file path
+                    audio_url = audio_temp_path
+
+                    # Create diarization input
+                    diarization_input = AudioDiarizationInput(
+                        audio_url=audio_url, topics=topics
+                    )
+
+                    # Run diarization
+                    await diarization_processor.push(diarization_input)
+                    await diarization_processor.flush()
+
+                    logger.info("Diarization complete")
+
+            except ImportError as e:
+                logger.error(f"Failed to import diarization dependencies: {e}")
+                logger.error(
+                    "Install with: uv pip install pyannote.audio torch torchaudio"
+                )
+                logger.error(
+                    "And set HF_TOKEN environment variable for pyannote models"
+                )
+                raise SystemExit(1)
+            except Exception as e:
+                logger.error(f"Diarization failed: {e}")
+                raise SystemExit(1)
+        else:
+            logger.warning("Skipping diarization: no topics available")
+
+    # Clean up temp file
+    if audio_temp_path:
+        try:
+            Path(audio_temp_path).unlink()
+        except Exception as e:
+            logger.warning(f"Failed to clean up temp file {audio_temp_path}: {e}")
+
+    logger.info("All done!")
+
+
+async def process_file_pipeline(
+    filename: str,
+    event_callback,
+    source_language="en",
+    target_language="en",
+    enable_diarization=True,
+    diarization_backend="modal",
+):
+    """Process audio/video file using the optimized file pipeline"""
+    try:
+        from reflector.db import database
+        from reflector.db.transcripts import SourceKind, transcripts_controller
+        from reflector.pipelines.main_file_pipeline import PipelineMainFile
+
+        await database.connect()
+        try:
+            # Create a temporary transcript for processing
+            transcript = await transcripts_controller.add(
+                "",
+                source_kind=SourceKind.FILE,
+                source_language=source_language,
+                target_language=target_language,
+            )
+
+            # Process the file
+            pipeline = PipelineMainFile(transcript_id=transcript.id)
+            await pipeline.process(Path(filename))
+
+            logger.info("File pipeline processing complete")
+
+        finally:
+            await database.disconnect()
+    except ImportError as e:
+        logger.error(f"File pipeline not available: {e}")
+        logger.info("Falling back to stream pipeline")
+        # Fall back to stream pipeline
+        await process_audio_file(
+            filename,
+            event_callback,
+            only_transcript=False,
+            source_language=source_language,
+            target_language=target_language,
+            enable_diarization=enable_diarization,
+            diarization_backend=diarization_backend,
+        )
 
 
 if __name__ == "__main__":
     import argparse
+    import os
 
-    parser = argparse.ArgumentParser()
+    parser = argparse.ArgumentParser(
+        description="Process audio files with optional speaker diarization"
+    )
     parser.add_argument("source", help="Source file (mp3, wav, mp4...)")
-    parser.add_argument("--only-transcript", "-t", action="store_true")
-    parser.add_argument("--source-language", default="en")
-    parser.add_argument("--target-language", default="en")
+    parser.add_argument(
+        "--stream",
+        action="store_true",
+        help="Use streaming pipeline (original frame-based processing)",
+    )
+    parser.add_argument(
+        "--only-transcript",
+        "-t",
+        action="store_true",
+        help="Only generate transcript without topics/summaries",
+    )
+    parser.add_argument(
+        "--source-language", default="en", help="Source language code (default: en)"
+    )
+    parser.add_argument(
+        "--target-language", default="en", help="Target language code (default: en)"
+    )
     parser.add_argument("--output", "-o", help="Output file (output.jsonl)")
+    parser.add_argument(
+        "--enable-diarization",
+        "-d",
+        action="store_true",
+        help="Enable speaker diarization",
+    )
+    parser.add_argument(
+        "--diarization-backend",
+        default="pyannote",
+        choices=["pyannote", "modal"],
+        help="Diarization backend to use (default: pyannote)",
+    )
     args = parser.parse_args()
 
+    if "REDIS_HOST" not in os.environ:
+        os.environ["REDIS_HOST"] = "localhost"
+
     output_fd = None
     if args.output:
         output_fd = open(args.output, "w")
 
     async def event_callback(event: PipelineEvent):
         processor = event.processor
-        # ignore some processor
-        if processor in ("AudioChunkerProcessor", "AudioMergeProcessor"):
+        data = event.data
+
+        # Ignore internal processors
+        if processor in (
+            "AudioDownscaleProcessor",
+            "AudioChunkerAutoProcessor",
+            "AudioMergeProcessor",
+            "AudioFileWriterProcessor",
+            "TopicCollectorProcessor",
+            "BroadcastProcessor",
+        ):
             return
-        logger.info(f"Event: {event}")
+
+        # If diarization is enabled, skip the original topic events from the pipeline
+        # The diarization processor will emit the same topics but with speaker info
+        if processor == "TranscriptTopicDetectorProcessor" and args.enable_diarization:
+            return
+
+        # Log all events
+        logger.info(f"Event: {processor} - {type(data).__name__}")
+
+        # Write to output
         if output_fd:
             output_fd.write(event.model_dump_json())
             output_fd.write("\n")
+            output_fd.flush()
 
+    if args.stream:
+        # Use original streaming pipeline
         asyncio.run(
             process_audio_file(
                 args.source,
@@ -97,6 +356,20 @@ if __name__ == "__main__":
                 only_transcript=args.only_transcript,
                 source_language=args.source_language,
                 target_language=args.target_language,
+                enable_diarization=args.enable_diarization,
+                diarization_backend=args.diarization_backend,
+            )
+        )
+    else:
+        # Use optimized file pipeline (default)
+        asyncio.run(
+            process_file_pipeline(
+                args.source,
+                event_callback,
+                source_language=args.source_language,
+                target_language=args.target_language,
+                enable_diarization=args.enable_diarization,
+                diarization_backend=args.diarization_backend,
             )
         )
 

server/reflector/tools/process_with_diarization.py

@@ -17,7 +17,8 @@ import av
 
 from reflector.logger import logger
 from reflector.processors import (
-    AudioChunkerProcessor,
+    AudioChunkerAutoProcessor,
+    AudioDownscaleProcessor,
     AudioFileWriterProcessor,
     AudioMergeProcessor,
     AudioTranscriptAutoProcessor,
@@ -96,7 +97,8 @@ async def process_audio_file_with_diarization(
 
     # Add the rest of the processors
     processors += [
-        AudioChunkerProcessor(),
+        AudioDownscaleProcessor(),
+        AudioChunkerAutoProcessor(),
         AudioMergeProcessor(),
         AudioTranscriptAutoProcessor.as_threaded(),
     ]
@@ -145,18 +147,17 @@ async def process_audio_file_with_diarization(
             logger.info(f"Starting diarization with {len(topics)} topics")
 
             try:
-                # Import diarization processor
                 from reflector.processors import AudioDiarizationAutoProcessor
 
-                # Create diarization processor
                 diarization_processor = AudioDiarizationAutoProcessor(
                     name=diarization_backend
                 )
-                diarization_processor.on(event_callback)
+
+                diarization_processor.set_pipeline(pipeline)
 
                 # For Modal backend, we need to upload the file to S3 first
                 if diarization_backend == "modal":
-                    from datetime import datetime
+                    from datetime import datetime, timezone
 
                     from reflector.storage import get_transcripts_storage
                     from reflector.utils.s3_temp_file import S3TemporaryFile
@@ -164,7 +165,7 @@ async def process_audio_file_with_diarization(
                     storage = get_transcripts_storage()
 
                     # Generate a unique filename in evaluation folder
-                    timestamp = datetime.utcnow().strftime("%Y%m%d_%H%M%S")
+                    timestamp = datetime.now(timezone.utc).strftime("%Y%m%d_%H%M%S")
                     audio_filename = f"evaluation/diarization_temp/{timestamp}_{uuid.uuid4().hex}.wav"
 
                     # Use context manager for automatic cleanup
@@ -277,7 +278,8 @@ if __name__ == "__main__":
 
         # Ignore internal processors
         if processor in (
-            "AudioChunkerProcessor",
+            "AudioDownscaleProcessor",
+            "AudioChunkerAutoProcessor",
             "AudioMergeProcessor",
             "AudioFileWriterProcessor",
             "TopicCollectorProcessor",

server/reflector/tools/runpipeline.py

@@ -53,7 +53,7 @@ async def run_single_processor(args):
     async def event_callback(event: PipelineEvent):
         processor = event.processor
         # ignore some processor
-        if processor in ("AudioChunkerProcessor", "AudioMergeProcessor"):
+        if processor in ("AudioChunkerAutoProcessor", "AudioMergeProcessor"):
             return
         print(f"Event: {event}")
         if output_fd:

server/reflector/utils/webvtt.py

@@ -0,0 +1,63 @@
+"""WebVTT utilities for generating subtitle files from transcript data."""
+
+from typing import TYPE_CHECKING, Annotated
+
+import webvtt
+
+from reflector.processors.types import Seconds, Word, words_to_segments
+
+if TYPE_CHECKING:
+    from reflector.db.transcripts import TranscriptTopic
+
+VttTimestamp = Annotated[str, "vtt_timestamp"]
+WebVTTStr = Annotated[str, "webvtt_str"]
+
+
+def _seconds_to_timestamp(seconds: Seconds) -> VttTimestamp:
+    # lib doesn't do that
+    hours = int(seconds // 3600)
+    minutes = int((seconds % 3600) // 60)
+    secs = int(seconds % 60)
+    milliseconds = int((seconds % 1) * 1000)
+
+    return f"{hours:02d}:{minutes:02d}:{secs:02d}.{milliseconds:03d}"
+
+
+def words_to_webvtt(words: list[Word]) -> WebVTTStr:
+    """Convert words to WebVTT using existing segmentation logic."""
+    vtt = webvtt.WebVTT()
+    if not words:
+        return vtt.content
+
+    segments = words_to_segments(words)
+
+    for segment in segments:
+        text = segment.text.strip()
+        # lib doesn't do that
+        text = f"<v Speaker{segment.speaker}>{text}"
+
+        caption = webvtt.Caption(
+            start=_seconds_to_timestamp(segment.start),
+            end=_seconds_to_timestamp(segment.end),
+            text=text,
+        )
+        vtt.captions.append(caption)
+
+    return vtt.content
+
+
+def topics_to_webvtt(topics: list["TranscriptTopic"]) -> WebVTTStr:
+    if not topics:
+        return webvtt.WebVTT().content
+
+    all_words: list[Word] = []
+    for topic in topics:
+        all_words.extend(topic.words)
+
+    # assert it's in sequence
+    for i in range(len(all_words) - 1):
+        assert (
+            all_words[i].start <= all_words[i + 1].start
+        ), f"Words are not in sequence: {all_words[i].text} and {all_words[i + 1].text} are not consecutive: {all_words[i].start} > {all_words[i + 1].start}"
+
+    return words_to_webvtt(all_words)

server/reflector/views/_range_requests_response.py

@@ -44,8 +44,6 @@ def range_requests_response(
     """Returns StreamingResponse using Range Requests of a given file"""
 
     if not os.path.exists(file_path):
-        from fastapi import HTTPException
-
         raise HTTPException(status_code=404, detail="File not found")
 
     file_size = os.stat(file_path).st_size

server/reflector/views/meetings.py

@@ -1,4 +1,4 @@
-from datetime import datetime
+from datetime import datetime, timezone
 from typing import Annotated, Optional
 
 from fastapi import APIRouter, Depends, HTTPException, Request
@@ -35,7 +35,7 @@ async def meeting_audio_consent(
         meeting_id=meeting_id,
         user_id=user_id,
         consent_given=request.consent_given,
-        consent_timestamp=datetime.utcnow(),
+        consent_timestamp=datetime.now(timezone.utc),
     )
 
     updated_consent = await meeting_consent_controller.upsert(consent)

server/reflector/views/rooms.py

@@ -1,16 +1,16 @@
 import logging
 import sqlite3
-from datetime import datetime, timedelta
+from datetime import datetime, timedelta, timezone
 from typing import Annotated, Literal, Optional
 
 import asyncpg.exceptions
 from fastapi import APIRouter, Depends, HTTPException
 from fastapi_pagination import Page
-from fastapi_pagination.ext.databases import paginate
+from fastapi_pagination.ext.databases import apaginate
 from pydantic import BaseModel
 
 import reflector.auth as auth
-from reflector.db import database
+from reflector.db import get_database
 from reflector.db.meetings import meetings_controller
 from reflector.db.rooms import rooms_controller
 from reflector.settings import settings
@@ -21,6 +21,14 @@ logger = logging.getLogger(__name__)
 router = APIRouter()
 
 
+def parse_datetime_with_timezone(iso_string: str) -> datetime:
+    """Parse ISO datetime string and ensure timezone awareness (defaults to UTC if naive)."""
+    dt = datetime.fromisoformat(iso_string)
+    if dt.tzinfo is None:
+        dt = dt.replace(tzinfo=timezone.utc)
+    return dt
+
+
 class Room(BaseModel):
     id: str
     name: str
@@ -83,8 +91,8 @@ async def rooms_list(
 
     user_id = user["sub"] if user else None
 
-    return await paginate(
-        database,
+    return await apaginate(
+        get_database(),
         await rooms_controller.get_all(
             user_id=user_id, order_by="-created_at", return_query=True
         ),
@@ -150,7 +158,7 @@ async def rooms_create_meeting(
     if not room:
         raise HTTPException(status_code=404, detail="Room not found")
 
-    current_time = datetime.utcnow()
+    current_time = datetime.now(timezone.utc)
     meeting = await meetings_controller.get_active(room=room, current_time=current_time)
 
     if meeting is None:
@@ -166,8 +174,8 @@ async def rooms_create_meeting(
                 room_name=whereby_meeting["roomName"],
                 room_url=whereby_meeting["roomUrl"],
                 host_room_url=whereby_meeting["hostRoomUrl"],
-                start_date=datetime.fromisoformat(whereby_meeting["startDate"]),
-                end_date=datetime.fromisoformat(whereby_meeting["endDate"]),
+                start_date=parse_datetime_with_timezone(whereby_meeting["startDate"]),
+                end_date=parse_datetime_with_timezone(whereby_meeting["endDate"]),
                 user_id=user_id,
                 room=room,
             )

server/reflector/views/transcripts.py

@@ -1,15 +1,29 @@
 from datetime import datetime, timedelta, timezone
 from typing import Annotated, Literal, Optional
 
-from fastapi import APIRouter, Depends, HTTPException
+from fastapi import APIRouter, Depends, HTTPException, Query
 from fastapi_pagination import Page
-from fastapi_pagination.ext.databases import paginate
+from fastapi_pagination.ext.databases import apaginate
 from jose import jwt
 from pydantic import BaseModel, Field, field_serializer
 
 import reflector.auth as auth
+from reflector.db import get_database
 from reflector.db.meetings import meetings_controller
 from reflector.db.rooms import rooms_controller
+from reflector.db.search import (
+    DEFAULT_SEARCH_LIMIT,
+    SearchLimit,
+    SearchLimitBase,
+    SearchOffset,
+    SearchOffsetBase,
+    SearchParameters,
+    SearchQuery,
+    SearchQueryBase,
+    SearchResult,
+    SearchTotal,
+    search_controller,
+)
 from reflector.db.transcripts import (
     SourceKind,
     TranscriptParticipant,
@@ -34,7 +48,7 @@ DOWNLOAD_EXPIRE_MINUTES = 60
 
 def create_access_token(data: dict, expires_delta: timedelta):
     to_encode = data.copy()
-    expire = datetime.utcnow() + expires_delta
+    expire = datetime.now(timezone.utc) + expires_delta
     to_encode.update({"exp": expire})
     encoded_jwt = jwt.encode(to_encode, settings.SECRET_KEY, algorithm=ALGORITHM)
     return encoded_jwt
@@ -100,6 +114,21 @@ class DeletionStatus(BaseModel):
     status: str
 
 
+SearchQueryParam = Annotated[SearchQueryBase, Query(description="Search query text")]
+SearchLimitParam = Annotated[SearchLimitBase, Query(description="Results per page")]
+SearchOffsetParam = Annotated[
+    SearchOffsetBase, Query(description="Number of results to skip")
+]
+
+
+class SearchResponse(BaseModel):
+    results: list[SearchResult]
+    total: SearchTotal
+    query: SearchQuery
+    limit: SearchLimit
+    offset: SearchOffset
+
+
 @router.get("/transcripts", response_model=Page[GetTranscriptMinimal])
 async def transcripts_list(
     user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
@@ -107,15 +136,13 @@ async def transcripts_list(
     room_id: str | None = None,
     search_term: str | None = None,
 ):
-    from reflector.db import database
-
     if not user and not settings.PUBLIC_MODE:
         raise HTTPException(status_code=401, detail="Not authenticated")
 
     user_id = user["sub"] if user else None
 
-    return await paginate(
-        database,
+    return await apaginate(
+        get_database(),
         await transcripts_controller.get_all(
             user_id=user_id,
             source_kind=SourceKind(source_kind) if source_kind else None,
@@ -127,6 +154,45 @@ async def transcripts_list(
     )
 
 
+@router.get("/transcripts/search", response_model=SearchResponse)
+async def transcripts_search(
+    q: SearchQueryParam,
+    limit: SearchLimitParam = DEFAULT_SEARCH_LIMIT,
+    offset: SearchOffsetParam = 0,
+    room_id: Optional[str] = None,
+    source_kind: Optional[SourceKind] = None,
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional)
+    ] = None,
+):
+    """
+    Full-text search across transcript titles and content.
+    """
+    if not user and not settings.PUBLIC_MODE:
+        raise HTTPException(status_code=401, detail="Not authenticated")
+
+    user_id = user["sub"] if user else None
+
+    search_params = SearchParameters(
+        query_text=q,
+        limit=limit,
+        offset=offset,
+        user_id=user_id,
+        room_id=room_id,
+        source_kind=source_kind,
+    )
+
+    results, total = await search_controller.search_transcripts(search_params)
+
+    return SearchResponse(
+        results=results,
+        total=total,
+        query=search_params.query_text,
+        limit=search_params.limit,
+        offset=search_params.offset,
+    )
+
+
 @router.post("/transcripts", response_model=GetTranscript)
 async def transcripts_create(
     info: CreateTranscript,
@@ -273,8 +339,8 @@ async def transcript_update(
     if not transcript:
         raise HTTPException(status_code=404, detail="Transcript not found")
     values = info.dict(exclude_unset=True)
-    await transcripts_controller.update(transcript, values)
-    return transcript
+    updated_transcript = await transcripts_controller.update(transcript, values)
+    return updated_transcript
 
 
 @router.delete("/transcripts/{transcript_id}", response_model=DeletionStatus)

server/reflector/views/transcripts_audio.py

@@ -51,24 +51,6 @@ async def transcript_get_audio_mp3(
         transcript_id, user_id=user_id
     )
 
-    if transcript.audio_location == "storage":
-        # proxy S3 file, to prevent issue with CORS
-        url = await transcript.get_audio_url()
-        headers = {}
-
-        copy_headers = ["range", "accept-encoding"]
-        for header in copy_headers:
-            if header in request.headers:
-                headers[header] = request.headers[header]
-
-        async with httpx.AsyncClient() as client:
-            resp = await client.request(request.method, url, headers=headers)
-            return Response(
-                content=resp.content,
-                status_code=resp.status_code,
-                headers=resp.headers,
-            )
-
     if transcript.audio_location == "storage":
         # proxy S3 file, to prevent issue with CORS
         url = await transcript.get_audio_url()

server/reflector/views/transcripts_webrtc.py

@@ -26,7 +26,7 @@ async def transcript_record_webrtc(
         raise HTTPException(status_code=400, detail="Transcript is locked")
 
     # create a pipeline runner
-    from reflector.pipelines.main_live_pipeline import PipelineMainLive
+    from reflector.pipelines.main_live_pipeline import PipelineMainLive  # noqa: PLC0415
 
     pipeline_runner = PipelineMainLive(transcript_id=transcript_id)
 

server/reflector/whereby.py

@@ -23,7 +23,7 @@ async def create_meeting(room_name_prefix: str, end_date: datetime, room: Room):
             "type": room.recording_type,
             "destination": {
                 "provider": "s3",
-                "bucket": settings.AWS_WHEREBY_S3_BUCKET,
+                "bucket": settings.RECORDING_STORAGE_AWS_BUCKET_NAME,
                 "accessKeyId": settings.AWS_WHEREBY_ACCESS_KEY_ID,
                 "accessKeySecret": settings.AWS_WHEREBY_ACCESS_KEY_SECRET,
                 "fileFormat": "mp4",

server/reflector/worker/process.py

@@ -14,13 +14,22 @@ from reflector.db.meetings import meetings_controller
 from reflector.db.recordings import Recording, recordings_controller
 from reflector.db.rooms import rooms_controller
 from reflector.db.transcripts import SourceKind, transcripts_controller
-from reflector.pipelines.main_live_pipeline import asynctask, task_pipeline_process
+from reflector.pipelines.main_file_pipeline import task_pipeline_file_process
+from reflector.pipelines.main_live_pipeline import asynctask
 from reflector.settings import settings
 from reflector.whereby import get_room_sessions
 
 logger = structlog.wrap_logger(get_task_logger(__name__))
 
 
+def parse_datetime_with_timezone(iso_string: str) -> datetime:
+    """Parse ISO datetime string and ensure timezone awareness (defaults to UTC if naive)."""
+    dt = datetime.fromisoformat(iso_string)
+    if dt.tzinfo is None:
+        dt = dt.replace(tzinfo=timezone.utc)
+    return dt
+
+
 @shared_task
 def process_messages():
     queue_url = settings.AWS_PROCESS_RECORDING_QUEUE_URL
@@ -69,7 +78,7 @@ async def process_recording(bucket_name: str, object_key: str):
 
     # extract a guid and a datetime from the object key
     room_name = f"/{object_key[:36]}"
-    recorded_at = datetime.fromisoformat(object_key[37:57])
+    recorded_at = parse_datetime_with_timezone(object_key[37:57])
 
     meeting = await meetings_controller.get_by_room_name(room_name)
     room = await rooms_controller.get_by_id(meeting.room_id)
@@ -132,7 +141,7 @@ async def process_recording(bucket_name: str, object_key: str):
 
     await transcripts_controller.update(transcript, {"status": "uploaded"})
 
-    task_pipeline_process.delay(transcript_id=transcript.id)
+    task_pipeline_file_process.delay(transcript_id=transcript.id)
 
 
 @shared_task
@@ -177,7 +186,7 @@ async def reprocess_failed_recordings():
     reprocessed_count = 0
     try:
         paginator = s3.get_paginator("list_objects_v2")
-        bucket_name = settings.AWS_WHEREBY_S3_BUCKET
+        bucket_name = settings.RECORDING_STORAGE_AWS_BUCKET_NAME
         pages = paginator.paginate(Bucket=bucket_name)
 
         for page in pages:

server/reflector/ws_manager.py

@@ -62,6 +62,7 @@ class RedisPubSubManager:
 class WebsocketManager:
     def __init__(self, pubsub_client: RedisPubSubManager = None):
         self.rooms: dict = {}
+        self.tasks: dict = {}
         self.pubsub_client = pubsub_client
 
     async def add_user_to_room(self, room_id: str, websocket: WebSocket) -> None:
@@ -74,13 +75,17 @@ class WebsocketManager:
 
             await self.pubsub_client.connect()
             pubsub_subscriber = await self.pubsub_client.subscribe(room_id)
-            asyncio.create_task(self._pubsub_data_reader(pubsub_subscriber))
+            task = asyncio.create_task(self._pubsub_data_reader(pubsub_subscriber))
+            self.tasks[id(websocket)] = task
 
     async def send_json(self, room_id: str, message: dict) -> None:
         await self.pubsub_client.send_json(room_id, message)
 
     async def remove_user_from_room(self, room_id: str, websocket: WebSocket) -> None:
         self.rooms[room_id].remove(websocket)
+        task = self.tasks.pop(id(websocket), None)
+        if task:
+            task.cancel()
 
         if len(self.rooms[room_id]) == 0:
             del self.rooms[room_id]

server/tests/cassettes/test_processors_modal/test_file_diarization_modal_processor.yaml

@@ -0,0 +1,40 @@
+interactions:
+- request:
+    body: ''
+    headers:
+      accept:
+      - '*/*'
+      accept-encoding:
+      - gzip, deflate
+      authorization:
+      - DUMMY_API_KEY
+      connection:
+      - keep-alive
+      content-length:
+      - '0'
+      host:
+      - monadical-sas--reflector-diarizer-web.modal.run
+      user-agent:
+      - python-httpx/0.27.2
+    method: POST
+    uri: https://monadical-sas--reflector-diarizer-web.modal.run/diarize?audio_file_url=https%3A%2F%2Freflector-github-pytest.s3.us-east-1.amazonaws.com%2Ftest_mathieu_hello.mp3&timestamp=0
+  response:
+    body:
+      string: '{"diarization":[{"start":0.823,"end":1.91,"speaker":0},{"start":2.572,"end":6.409,"speaker":0},{"start":6.783,"end":10.62,"speaker":0},{"start":11.231,"end":14.168,"speaker":0},{"start":14.796,"end":19.295,"speaker":0}]}'
+    headers:
+      Alt-Svc:
+      - h3=":443"; ma=2592000
+      Content-Length:
+      - '220'
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 13 Aug 2025 18:25:34 GMT
+      Modal-Function-Call-Id:
+      - fc-01K2JAVNEP6N7Y1Y7W3T98BCXK
+      Vary:
+      - accept-encoding
+    status:
+      code: 200
+      message: OK
+version: 1

server/tests/cassettes/test_processors_modal/test_file_transcript_modal_processor.yaml

@@ -0,0 +1,46 @@
+interactions:
+- request:
+    body: '{"audio_file_url": "https://reflector-github-pytest.s3.us-east-1.amazonaws.com/test_mathieu_hello.mp3",
+      "language": "en", "batch": true}'
+    headers:
+      accept:
+      - '*/*'
+      accept-encoding:
+      - gzip, deflate
+      authorization:
+      - DUMMY_API_KEY
+      connection:
+      - keep-alive
+      content-length:
+      - '136'
+      content-type:
+      - application/json
+      host:
+      - monadical-sas--reflector-transcriber-parakeet-web.modal.run
+      user-agent:
+      - python-httpx/0.27.2
+    method: POST
+    uri: https://monadical-sas--reflector-transcriber-parakeet-web.modal.run/v1/audio/transcriptions-from-url
+  response:
+    body:
+      string: '{"text":"Hi there everyone. Today I want to share my incredible experience
+        with Reflector. a Q teenage product that revolutionizes audio processing.
+        With reflector, I can easily convert any audio into accurate transcription.
+        saving me hours of tedious manual work.","words":[{"word":"Hi","start":0.87,"end":1.19},{"word":"there","start":1.19,"end":1.35},{"word":"everyone.","start":1.51,"end":1.83},{"word":"Today","start":2.63,"end":2.87},{"word":"I","start":3.36,"end":3.52},{"word":"want","start":3.6,"end":3.76},{"word":"to","start":3.76,"end":3.92},{"word":"share","start":3.92,"end":4.16},{"word":"my","start":4.16,"end":4.4},{"word":"incredible","start":4.32,"end":4.96},{"word":"experience","start":4.96,"end":5.44},{"word":"with","start":5.44,"end":5.68},{"word":"Reflector.","start":5.68,"end":6.24},{"word":"a","start":6.93,"end":7.01},{"word":"Q","start":7.01,"end":7.17},{"word":"teenage","start":7.25,"end":7.65},{"word":"product","start":7.89,"end":8.29},{"word":"that","start":8.29,"end":8.61},{"word":"revolutionizes","start":8.61,"end":9.65},{"word":"audio","start":9.65,"end":10.05},{"word":"processing.","start":10.05,"end":10.53},{"word":"With","start":11.27,"end":11.43},{"word":"reflector,","start":11.51,"end":12.15},{"word":"I","start":12.31,"end":12.39},{"word":"can","start":12.39,"end":12.55},{"word":"easily","start":12.55,"end":12.95},{"word":"convert","start":12.95,"end":13.43},{"word":"any","start":13.43,"end":13.67},{"word":"audio","start":13.67,"end":13.99},{"word":"into","start":14.98,"end":15.06},{"word":"accurate","start":15.22,"end":15.54},{"word":"transcription.","start":15.7,"end":16.34},{"word":"saving","start":16.99,"end":17.15},{"word":"me","start":17.31,"end":17.47},{"word":"hours","start":17.47,"end":17.87},{"word":"of","start":17.87,"end":18.11},{"word":"tedious","start":18.11,"end":18.67},{"word":"manual","start":18.67,"end":19.07},{"word":"work.","start":19.07,"end":19.31}]}'
+    headers:
+      Alt-Svc:
+      - h3=":443"; ma=2592000
+      Content-Length:
+      - '1933'
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 13 Aug 2025 18:26:59 GMT
+      Modal-Function-Call-Id:
+      - fc-01K2JAWC7GAMKX4DSJ21WV31NG
+      Vary:
+      - accept-encoding
+    status:
+      code: 200
+      message: OK
+version: 1

server/tests/cassettes/test_processors_modal/test_full_modal_pipeline_integration.yaml

@@ -0,0 +1,84 @@
+interactions:
+- request:
+    body: '{"audio_file_url": "https://reflector-github-pytest.s3.us-east-1.amazonaws.com/test_mathieu_hello.mp3",
+      "language": "en", "batch": true}'
+    headers:
+      accept:
+      - '*/*'
+      accept-encoding:
+      - gzip, deflate
+      authorization:
+      - DUMMY_API_KEY
+      connection:
+      - keep-alive
+      content-length:
+      - '136'
+      content-type:
+      - application/json
+      host:
+      - monadical-sas--reflector-transcriber-parakeet-web.modal.run
+      user-agent:
+      - python-httpx/0.27.2
+    method: POST
+    uri: https://monadical-sas--reflector-transcriber-parakeet-web.modal.run/v1/audio/transcriptions-from-url
+  response:
+    body:
+      string: '{"text":"Hi there everyone. Today I want to share my incredible experience
+        with Reflector. a Q teenage product that revolutionizes audio processing.
+        With reflector, I can easily convert any audio into accurate transcription.
+        saving me hours of tedious manual work.","words":[{"word":"Hi","start":0.87,"end":1.19},{"word":"there","start":1.19,"end":1.35},{"word":"everyone.","start":1.51,"end":1.83},{"word":"Today","start":2.63,"end":2.87},{"word":"I","start":3.36,"end":3.52},{"word":"want","start":3.6,"end":3.76},{"word":"to","start":3.76,"end":3.92},{"word":"share","start":3.92,"end":4.16},{"word":"my","start":4.16,"end":4.4},{"word":"incredible","start":4.32,"end":4.96},{"word":"experience","start":4.96,"end":5.44},{"word":"with","start":5.44,"end":5.68},{"word":"Reflector.","start":5.68,"end":6.24},{"word":"a","start":6.93,"end":7.01},{"word":"Q","start":7.01,"end":7.17},{"word":"teenage","start":7.25,"end":7.65},{"word":"product","start":7.89,"end":8.29},{"word":"that","start":8.29,"end":8.61},{"word":"revolutionizes","start":8.61,"end":9.65},{"word":"audio","start":9.65,"end":10.05},{"word":"processing.","start":10.05,"end":10.53},{"word":"With","start":11.27,"end":11.43},{"word":"reflector,","start":11.51,"end":12.15},{"word":"I","start":12.31,"end":12.39},{"word":"can","start":12.39,"end":12.55},{"word":"easily","start":12.55,"end":12.95},{"word":"convert","start":12.95,"end":13.43},{"word":"any","start":13.43,"end":13.67},{"word":"audio","start":13.67,"end":13.99},{"word":"into","start":14.98,"end":15.06},{"word":"accurate","start":15.22,"end":15.54},{"word":"transcription.","start":15.7,"end":16.34},{"word":"saving","start":16.99,"end":17.15},{"word":"me","start":17.31,"end":17.47},{"word":"hours","start":17.47,"end":17.87},{"word":"of","start":17.87,"end":18.11},{"word":"tedious","start":18.11,"end":18.67},{"word":"manual","start":18.67,"end":19.07},{"word":"work.","start":19.07,"end":19.31}]}'
+    headers:
+      Alt-Svc:
+      - h3=":443"; ma=2592000
+      Content-Length:
+      - '1933'
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 13 Aug 2025 18:27:02 GMT
+      Modal-Function-Call-Id:
+      - fc-01K2JAYZ1AR2HE422VJVKBWX9Z
+      Vary:
+      - accept-encoding
+    status:
+      code: 200
+      message: OK
+- request:
+    body: ''
+    headers:
+      accept:
+      - '*/*'
+      accept-encoding:
+      - gzip, deflate
+      authorization:
+      - DUMMY_API_KEY
+      connection:
+      - keep-alive
+      content-length:
+      - '0'
+      host:
+      - monadical-sas--reflector-diarizer-web.modal.run
+      user-agent:
+      - python-httpx/0.27.2
+    method: POST
+    uri: https://monadical-sas--reflector-diarizer-web.modal.run/diarize?audio_file_url=https%3A%2F%2Freflector-github-pytest.s3.us-east-1.amazonaws.com%2Ftest_mathieu_hello.mp3&timestamp=0
+  response:
+    body:
+      string: '{"diarization":[{"start":0.823,"end":1.91,"speaker":0},{"start":2.572,"end":6.409,"speaker":0},{"start":6.783,"end":10.62,"speaker":0},{"start":11.231,"end":14.168,"speaker":0},{"start":14.796,"end":19.295,"speaker":0}]}'
+    headers:
+      Alt-Svc:
+      - h3=":443"; ma=2592000
+      Content-Length:
+      - '220'
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 13 Aug 2025 18:27:18 GMT
+      Modal-Function-Call-Id:
+      - fc-01K2JAZ1M34NQRJK03CCFK95D6
+      Vary:
+      - accept-encoding
+    status:
+      code: 200
+      message: OK
+version: 1

server/tests/conftest.py

@@ -1,17 +1,85 @@
+import os
 from tempfile import NamedTemporaryFile
 from unittest.mock import patch
 
 import pytest
 
 
+@pytest.fixture(scope="session", autouse=True)
+def settings_configuration():
+    # theses settings are linked to monadical for pytest-recording
+    # if a fork is done, they have to provide their own url when cassettes needs to be updated
+    # modal api keys has to be defined by the user
+    from reflector.settings import settings
+
+    settings.TRANSCRIPT_BACKEND = "modal"
+    settings.TRANSCRIPT_URL = (
+        "https://monadical-sas--reflector-transcriber-parakeet-web.modal.run"
+    )
+    settings.DIARIZATION_BACKEND = "modal"
+    settings.DIARIZATION_URL = "https://monadical-sas--reflector-diarizer-web.modal.run"
+
+
+@pytest.fixture(scope="module")
+def vcr_config():
+    """VCR configuration to filter sensitive headers"""
+    return {
+        "filter_headers": [("authorization", "DUMMY_API_KEY")],
+    }
+
+
+@pytest.fixture(scope="session")
+def docker_compose_file(pytestconfig):
+    return os.path.join(str(pytestconfig.rootdir), "tests", "docker-compose.test.yml")
+
+
+@pytest.fixture(scope="session")
+def postgres_service(docker_ip, docker_services):
+    """Ensure that PostgreSQL service is up and responsive."""
+    port = docker_services.port_for("postgres_test", 5432)
+
+    def is_responsive():
+        try:
+            import psycopg2
+
+            conn = psycopg2.connect(
+                host=docker_ip,
+                port=port,
+                dbname="reflector_test",
+                user="test_user",
+                password="test_password",
+            )
+            conn.close()
+            return True
+        except Exception:
+            return False
+
+    docker_services.wait_until_responsive(timeout=30.0, pause=0.1, check=is_responsive)
+
+    # Return connection parameters
+    return {
+        "host": docker_ip,
+        "port": port,
+        "dbname": "reflector_test",
+        "user": "test_user",
+        "password": "test_password",
+    }
+
+
 @pytest.fixture(scope="function", autouse=True)
 @pytest.mark.asyncio
-async def setup_database():
-    from reflector.db import engine, metadata  # noqa
+async def setup_database(postgres_service):
+    from reflector.db import engine, metadata, get_database  # noqa
 
     metadata.drop_all(bind=engine)
     metadata.create_all(bind=engine)
+    database = get_database()
+
+    try:
+        await database.connect()
         yield
+    finally:
+        await database.disconnect()
 
 
 @pytest.fixture
@@ -46,6 +114,20 @@ def dummy_processors():
         )  # noqa
 
 
+@pytest.fixture
+async def whisper_transcript():
+    from reflector.processors.audio_transcript_whisper import (
+        AudioTranscriptWhisperProcessor,
+    )
+
+    with patch(
+        "reflector.processors.audio_transcript_auto"
+        ".AudioTranscriptAutoProcessor.__new__"
+    ) as mock_audio:
+        mock_audio.return_value = AudioTranscriptWhisperProcessor()
+        yield
+
+
 @pytest.fixture
 async def dummy_transcript():
     from reflector.processors.audio_transcript import AudioTranscriptProcessor
@@ -181,6 +263,16 @@ def celery_includes():
     return ["reflector.pipelines.main_live_pipeline"]
 
 
+@pytest.fixture
+async def client():
+    from httpx import AsyncClient
+
+    from reflector.app import app
+
+    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
+        yield ac
+
+
 @pytest.fixture(scope="session")
 def fake_mp3_upload():
     with patch(
@@ -191,13 +283,10 @@ def fake_mp3_upload():
 
 
 @pytest.fixture
-async def fake_transcript_with_topics(tmpdir):
+async def fake_transcript_with_topics(tmpdir, client):
     import shutil
     from pathlib import Path
 
-    from httpx import AsyncClient
-
-    from reflector.app import app
     from reflector.db.transcripts import TranscriptTopic
     from reflector.processors.types import Word
     from reflector.settings import settings
@@ -206,8 +295,7 @@ async def fake_transcript_with_topics(tmpdir):
     settings.DATA_DIR = Path(tmpdir)
 
     # create a transcript
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.post("/transcripts", json={"name": "Test audio download"})
+    response = await client.post("/transcripts", json={"name": "Test audio download"})
     assert response.status_code == 200
     tid = response.json()["id"]
 

server/tests/docker-compose.test.yml

@@ -0,0 +1,13 @@
+version: "3.8"
+services:
+  postgres_test:
+    image: postgres:17
+    environment:
+      POSTGRES_DB: reflector_test
+      POSTGRES_USER: test_user
+      POSTGRES_PASSWORD: test_password
+    ports:
+      - "15432:5432"
+    command: postgres -c fsync=off -c synchronous_commit=off -c full_page_writes=off
+    tmpfs:
+      - /var/lib/postgresql/data:rw,noexec,nosuid,size=1g

server/tests/test_gpu_modal_transcript.py

@@ -0,0 +1,330 @@
+"""
+Tests for GPU Modal transcription endpoints.
+
+These tests are marked with the "gpu-modal" group and will not run by default.
+Run them with: pytest -m gpu-modal tests/test_gpu_modal_transcript_parakeet.py
+
+Required environment variables:
+- TRANSCRIPT_URL: URL to the Modal.com endpoint (required)
+- TRANSCRIPT_MODAL_API_KEY: API key for authentication (optional)
+- TRANSCRIPT_MODEL: Model name to use (optional, defaults to nvidia/parakeet-tdt-0.6b-v2)
+
+Example with pytest (override default addopts to run ONLY gpu_modal tests):
+    TRANSCRIPT_URL=https://monadical-sas--reflector-transcriber-parakeet-web-dev.modal.run \
+    TRANSCRIPT_MODAL_API_KEY=your-api-key \
+    uv run -m pytest -m gpu_modal --no-cov tests/test_gpu_modal_transcript.py
+
+    # Or with completely clean options:
+    uv run -m pytest -m gpu_modal -o addopts="" tests/
+
+Running Modal locally for testing:
+    modal serve gpu/modal_deployments/reflector_transcriber_parakeet.py
+    # This will give you a local URL like https://xxxxx--reflector-transcriber-parakeet-web-dev.modal.run to test against
+"""
+
+import os
+import tempfile
+from pathlib import Path
+
+import httpx
+import pytest
+
+# Test audio file URL for testing
+TEST_AUDIO_URL = (
+    "https://reflector-github-pytest.s3.us-east-1.amazonaws.com/test_mathieu_hello.mp3"
+)
+
+
+def get_modal_transcript_url():
+    """Get and validate the Modal transcript URL from environment."""
+    url = os.environ.get("TRANSCRIPT_URL")
+    if not url:
+        pytest.skip(
+            "TRANSCRIPT_URL environment variable is required for GPU Modal tests"
+        )
+    return url
+
+
+def get_auth_headers():
+    """Get authentication headers if API key is available."""
+    api_key = os.environ.get("TRANSCRIPT_MODAL_API_KEY")
+    if api_key:
+        return {"Authorization": f"Bearer {api_key}"}
+    return {}
+
+
+def get_model_name():
+    """Get the model name from environment or use default."""
+    return os.environ.get("TRANSCRIPT_MODEL", "nvidia/parakeet-tdt-0.6b-v2")
+
+
+@pytest.mark.gpu_modal
+class TestGPUModalTranscript:
+    """Test suite for GPU Modal transcription endpoints."""
+
+    def test_transcriptions_from_url(self):
+        """Test the /v1/audio/transcriptions-from-url endpoint."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        with httpx.Client(timeout=60.0) as client:
+            response = client.post(
+                f"{url}/v1/audio/transcriptions-from-url",
+                json={
+                    "audio_file_url": TEST_AUDIO_URL,
+                    "model": get_model_name(),
+                    "language": "en",
+                    "timestamp_offset": 0.0,
+                },
+                headers=headers,
+            )
+
+            assert response.status_code == 200, f"Request failed: {response.text}"
+            result = response.json()
+
+            # Verify response structure
+            assert "text" in result
+            assert "words" in result
+            assert isinstance(result["text"], str)
+            assert isinstance(result["words"], list)
+
+            # Verify content is meaningful
+            assert len(result["text"]) > 0, "Transcript text should not be empty"
+            assert len(result["words"]) > 0, "Words list must not be empty"
+
+            # Verify word structure
+            for word in result["words"]:
+                assert "word" in word
+                assert "start" in word
+                assert "end" in word
+                assert isinstance(word["start"], (int, float))
+                assert isinstance(word["end"], (int, float))
+                assert word["start"] <= word["end"]
+
+    def test_transcriptions_single_file(self):
+        """Test the /v1/audio/transcriptions endpoint with a single file."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        # Download test audio file to upload
+        with httpx.Client(timeout=60.0) as client:
+            audio_response = client.get(TEST_AUDIO_URL)
+            audio_response.raise_for_status()
+
+            with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp_file:
+                tmp_file.write(audio_response.content)
+                tmp_file_path = tmp_file.name
+
+            try:
+                # Upload the file for transcription
+                with open(tmp_file_path, "rb") as f:
+                    files = {"file": ("test_audio.mp3", f, "audio/mpeg")}
+                    data = {
+                        "model": get_model_name(),
+                        "language": "en",
+                        "batch": "false",
+                    }
+
+                    response = client.post(
+                        f"{url}/v1/audio/transcriptions",
+                        files=files,
+                        data=data,
+                        headers=headers,
+                    )
+
+                assert response.status_code == 200, f"Request failed: {response.text}"
+                result = response.json()
+
+                # Verify response structure for single file
+                assert "text" in result
+                assert "words" in result
+                assert "filename" in result
+                assert isinstance(result["text"], str)
+                assert isinstance(result["words"], list)
+
+                # Verify content
+                assert len(result["text"]) > 0, "Transcript text should not be empty"
+
+            finally:
+                Path(tmp_file_path).unlink(missing_ok=True)
+
+    def test_transcriptions_multiple_files(self):
+        """Test the /v1/audio/transcriptions endpoint with multiple files (non-batch mode)."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        # Create multiple test files (we'll use the same audio content for simplicity)
+        with httpx.Client(timeout=60.0) as client:
+            audio_response = client.get(TEST_AUDIO_URL)
+            audio_response.raise_for_status()
+            audio_content = audio_response.content
+
+            temp_files = []
+            try:
+                # Create 3 temporary files
+                for i in range(3):
+                    tmp_file = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
+                    tmp_file.write(audio_content)
+                    tmp_file.close()
+                    temp_files.append(tmp_file.name)
+
+                # Upload multiple files for transcription (non-batch)
+                files = [
+                    ("files", (f"test_audio_{i}.mp3", open(f, "rb"), "audio/mpeg"))
+                    for i, f in enumerate(temp_files)
+                ]
+                data = {
+                    "model": get_model_name(),
+                    "language": "en",
+                    "batch": "false",
+                }
+
+                response = client.post(
+                    f"{url}/v1/audio/transcriptions",
+                    files=files,
+                    data=data,
+                    headers=headers,
+                )
+
+                # Close file handles
+                for _, file_tuple in files:
+                    file_tuple[1].close()
+
+                assert response.status_code == 200, f"Request failed: {response.text}"
+                result = response.json()
+
+                # Verify response structure for multiple files (non-batch)
+                assert "results" in result
+                assert isinstance(result["results"], list)
+                assert len(result["results"]) == 3
+
+                for idx, file_result in enumerate(result["results"]):
+                    assert "text" in file_result
+                    assert "words" in file_result
+                    assert "filename" in file_result
+                    assert isinstance(file_result["text"], str)
+                    assert isinstance(file_result["words"], list)
+                    assert len(file_result["text"]) > 0
+
+            finally:
+                for f in temp_files:
+                    Path(f).unlink(missing_ok=True)
+
+    def test_transcriptions_multiple_files_batch(self):
+        """Test the /v1/audio/transcriptions endpoint with multiple files in batch mode."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        # Create multiple test files
+        with httpx.Client(timeout=60.0) as client:
+            audio_response = client.get(TEST_AUDIO_URL)
+            audio_response.raise_for_status()
+            audio_content = audio_response.content
+
+            temp_files = []
+            try:
+                # Create 3 temporary files
+                for i in range(3):
+                    tmp_file = tempfile.NamedTemporaryFile(suffix=".mp3", delete=False)
+                    tmp_file.write(audio_content)
+                    tmp_file.close()
+                    temp_files.append(tmp_file.name)
+
+                # Upload multiple files for batch transcription
+                files = [
+                    ("files", (f"test_audio_{i}.mp3", open(f, "rb"), "audio/mpeg"))
+                    for i, f in enumerate(temp_files)
+                ]
+                data = {
+                    "model": get_model_name(),
+                    "language": "en",
+                    "batch": "true",
+                }
+
+                response = client.post(
+                    f"{url}/v1/audio/transcriptions",
+                    files=files,
+                    data=data,
+                    headers=headers,
+                )
+
+                # Close file handles
+                for _, file_tuple in files:
+                    file_tuple[1].close()
+
+                assert response.status_code == 200, f"Request failed: {response.text}"
+                result = response.json()
+
+                # Verify response structure for batch mode
+                assert "results" in result
+                assert isinstance(result["results"], list)
+                assert len(result["results"]) == 3
+
+                for idx, batch_result in enumerate(result["results"]):
+                    assert "text" in batch_result
+                    assert "words" in batch_result
+                    assert "filename" in batch_result
+                    assert isinstance(batch_result["text"], str)
+                    assert isinstance(batch_result["words"], list)
+                    assert len(batch_result["text"]) > 0
+
+            finally:
+                for f in temp_files:
+                    Path(f).unlink(missing_ok=True)
+
+    def test_transcriptions_error_handling(self):
+        """Test error handling for invalid requests."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        with httpx.Client(timeout=60.0) as client:
+            # Test with unsupported language
+            response = client.post(
+                f"{url}/v1/audio/transcriptions-from-url",
+                json={
+                    "audio_file_url": TEST_AUDIO_URL,
+                    "model": get_model_name(),
+                    "language": "fr",  # Parakeet only supports English
+                    "timestamp_offset": 0.0,
+                },
+                headers=headers,
+            )
+
+            assert response.status_code == 400
+            assert "only supports English" in response.text
+
+    def test_transcriptions_with_timestamp_offset(self):
+        """Test transcription with timestamp offset parameter."""
+        url = get_modal_transcript_url()
+        headers = get_auth_headers()
+
+        with httpx.Client(timeout=60.0) as client:
+            # Test with timestamp offset
+            response = client.post(
+                f"{url}/v1/audio/transcriptions-from-url",
+                json={
+                    "audio_file_url": TEST_AUDIO_URL,
+                    "model": get_model_name(),
+                    "language": "en",
+                    "timestamp_offset": 10.0,  # Add 10 second offset
+                },
+                headers=headers,
+            )
+
+            assert response.status_code == 200, f"Request failed: {response.text}"
+            result = response.json()
+
+            # Verify response structure
+            assert "text" in result
+            assert "words" in result
+            assert len(result["words"]) > 0, "Words list must not be empty"
+
+            # Verify that timestamps have been offset
+            for word in result["words"]:
+                # All timestamps should be >= 10.0 due to offset
+                assert (
+                    word["start"] >= 10.0
+                ), f"Word start time {word['start']} should be >= 10.0"
+                assert (
+                    word["end"] >= 10.0
+                ), f"Word end time {word['end']} should be >= 10.0"

server/tests/test_pipeline_main_file.py

@@ -0,0 +1,633 @@
+"""
+Tests for PipelineMainFile - file-based processing pipeline
+
+This test verifies the complete file processing pipeline without mocking much,
+ensuring all processors are correctly invoked and the happy path works correctly.
+"""
+
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock, patch
+from uuid import uuid4
+
+import pytest
+
+from reflector.pipelines.main_file_pipeline import PipelineMainFile
+from reflector.processors.file_diarization import FileDiarizationOutput
+from reflector.processors.types import (
+    DiarizationSegment,
+    TitleSummary,
+    Word,
+)
+from reflector.processors.types import (
+    Transcript as TranscriptType,
+)
+
+
+@pytest.fixture
+async def dummy_file_transcript():
+    """Mock FileTranscriptAutoProcessor for file processing"""
+    from reflector.processors.file_transcript import FileTranscriptProcessor
+
+    class TestFileTranscriptProcessor(FileTranscriptProcessor):
+        async def _transcript(self, data):
+            return TranscriptType(
+                text="Hello world. How are you today?",
+                words=[
+                    Word(start=0.0, end=0.5, text="Hello", speaker=0),
+                    Word(start=0.5, end=0.6, text=" ", speaker=0),
+                    Word(start=0.6, end=1.0, text="world", speaker=0),
+                    Word(start=1.0, end=1.1, text=".", speaker=0),
+                    Word(start=1.1, end=1.2, text=" ", speaker=0),
+                    Word(start=1.2, end=1.5, text="How", speaker=0),
+                    Word(start=1.5, end=1.6, text=" ", speaker=0),
+                    Word(start=1.6, end=1.8, text="are", speaker=0),
+                    Word(start=1.8, end=1.9, text=" ", speaker=0),
+                    Word(start=1.9, end=2.1, text="you", speaker=0),
+                    Word(start=2.1, end=2.2, text=" ", speaker=0),
+                    Word(start=2.2, end=2.5, text="today", speaker=0),
+                    Word(start=2.5, end=2.6, text="?", speaker=0),
+                ],
+            )
+
+    with patch(
+        "reflector.processors.file_transcript_auto.FileTranscriptAutoProcessor.__new__"
+    ) as mock_auto:
+        mock_auto.return_value = TestFileTranscriptProcessor()
+        yield
+
+
+@pytest.fixture
+async def dummy_file_diarization():
+    """Mock FileDiarizationAutoProcessor for file processing"""
+    from reflector.processors.file_diarization import FileDiarizationProcessor
+
+    class TestFileDiarizationProcessor(FileDiarizationProcessor):
+        async def _diarize(self, data):
+            return FileDiarizationOutput(
+                diarization=[
+                    DiarizationSegment(start=0.0, end=1.1, speaker=0),
+                    DiarizationSegment(start=1.2, end=2.6, speaker=1),
+                ]
+            )
+
+    with patch(
+        "reflector.processors.file_diarization_auto.FileDiarizationAutoProcessor.__new__"
+    ) as mock_auto:
+        mock_auto.return_value = TestFileDiarizationProcessor()
+        yield
+
+
+@pytest.fixture
+async def mock_transcript_in_db(tmpdir):
+    """Create a mock transcript in the database"""
+    from reflector.db.transcripts import Transcript
+    from reflector.settings import settings
+
+    # Set the DATA_DIR to our tmpdir
+    original_data_dir = settings.DATA_DIR
+    settings.DATA_DIR = str(tmpdir)
+
+    transcript_id = str(uuid4())
+    data_path = Path(tmpdir) / transcript_id
+    data_path.mkdir(parents=True, exist_ok=True)
+
+    # Create mock transcript object
+    transcript = Transcript(
+        id=transcript_id,
+        name="Test Transcript",
+        status="processing",
+        source_kind="file",
+        source_language="en",
+        target_language="en",
+    )
+
+    # Mock the controller to return our transcript
+    try:
+        with patch(
+            "reflector.pipelines.main_file_pipeline.transcripts_controller.get_by_id"
+        ) as mock_get:
+            mock_get.return_value = transcript
+            with patch(
+                "reflector.pipelines.main_live_pipeline.transcripts_controller.get_by_id"
+            ) as mock_get2:
+                mock_get2.return_value = transcript
+                with patch(
+                    "reflector.pipelines.main_live_pipeline.transcripts_controller.update"
+                ) as mock_update:
+                    mock_update.return_value = None
+                    yield transcript
+    finally:
+        # Restore original DATA_DIR
+        settings.DATA_DIR = original_data_dir
+
+
+@pytest.fixture
+async def mock_storage():
+    """Mock storage for file uploads"""
+    from reflector.storage.base import Storage
+
+    class TestStorage(Storage):
+        async def _put_file(self, path, data):
+            return None
+
+        async def _get_file_url(self, path):
+            return f"http://test-storage/{path}"
+
+        async def _get_file(self, path):
+            return b"test_audio_data"
+
+        async def _delete_file(self, path):
+            return None
+
+    storage = TestStorage()
+    # Add mock tracking for verification
+    storage._put_file = AsyncMock(side_effect=storage._put_file)
+    storage._get_file_url = AsyncMock(side_effect=storage._get_file_url)
+
+    with patch(
+        "reflector.pipelines.main_file_pipeline.get_transcripts_storage"
+    ) as mock_get:
+        mock_get.return_value = storage
+        yield storage
+
+
+@pytest.fixture
+async def mock_audio_file_writer():
+    """Mock AudioFileWriterProcessor to avoid actual file writing"""
+    with patch(
+        "reflector.pipelines.main_file_pipeline.AudioFileWriterProcessor"
+    ) as mock_writer_class:
+        mock_writer = AsyncMock()
+        mock_writer.push = AsyncMock()
+        mock_writer.flush = AsyncMock()
+        mock_writer_class.return_value = mock_writer
+        yield mock_writer
+
+
+@pytest.fixture
+async def mock_waveform_processor():
+    """Mock AudioWaveformProcessor"""
+    with patch(
+        "reflector.pipelines.main_file_pipeline.AudioWaveformProcessor"
+    ) as mock_waveform_class:
+        mock_waveform = AsyncMock()
+        mock_waveform.set_pipeline = MagicMock()
+        mock_waveform.flush = AsyncMock()
+        mock_waveform_class.return_value = mock_waveform
+        yield mock_waveform
+
+
+@pytest.fixture
+async def mock_topic_detector():
+    """Mock TranscriptTopicDetectorProcessor"""
+    with patch(
+        "reflector.pipelines.main_file_pipeline.TranscriptTopicDetectorProcessor"
+    ) as mock_topic_class:
+        mock_topic = AsyncMock()
+        mock_topic.set_pipeline = MagicMock()
+        mock_topic.push = AsyncMock()
+        mock_topic.flush_called = False
+
+        # When flush is called, simulate topic detection by calling the callback
+        async def flush_with_callback():
+            mock_topic.flush_called = True
+            if hasattr(mock_topic, "_callback"):
+                # Create a minimal transcript for the TitleSummary
+                test_transcript = TranscriptType(words=[], text="test transcript")
+                await mock_topic._callback(
+                    TitleSummary(
+                        title="Test Topic",
+                        summary="Test topic summary",
+                        timestamp=0.0,
+                        duration=10.0,
+                        transcript=test_transcript,
+                    )
+                )
+
+        mock_topic.flush = flush_with_callback
+
+        def init_with_callback(callback=None):
+            mock_topic._callback = callback
+            return mock_topic
+
+        mock_topic_class.side_effect = init_with_callback
+        yield mock_topic
+
+
+@pytest.fixture
+async def mock_title_processor():
+    """Mock TranscriptFinalTitleProcessor"""
+    with patch(
+        "reflector.pipelines.main_file_pipeline.TranscriptFinalTitleProcessor"
+    ) as mock_title_class:
+        mock_title = AsyncMock()
+        mock_title.set_pipeline = MagicMock()
+        mock_title.push = AsyncMock()
+        mock_title.flush_called = False
+
+        # When flush is called, simulate title generation by calling the callback
+        async def flush_with_callback():
+            mock_title.flush_called = True
+            if hasattr(mock_title, "_callback"):
+                from reflector.processors.types import FinalTitle
+
+                await mock_title._callback(FinalTitle(title="Test Title"))
+
+        mock_title.flush = flush_with_callback
+
+        def init_with_callback(callback=None):
+            mock_title._callback = callback
+            return mock_title
+
+        mock_title_class.side_effect = init_with_callback
+        yield mock_title
+
+
+@pytest.fixture
+async def mock_summary_processor():
+    """Mock TranscriptFinalSummaryProcessor"""
+    with patch(
+        "reflector.pipelines.main_file_pipeline.TranscriptFinalSummaryProcessor"
+    ) as mock_summary_class:
+        mock_summary = AsyncMock()
+        mock_summary.set_pipeline = MagicMock()
+        mock_summary.push = AsyncMock()
+        mock_summary.flush_called = False
+
+        # When flush is called, simulate summary generation by calling the callbacks
+        async def flush_with_callback():
+            mock_summary.flush_called = True
+            from reflector.processors.types import FinalLongSummary, FinalShortSummary
+
+            if hasattr(mock_summary, "_callback"):
+                await mock_summary._callback(
+                    FinalLongSummary(long_summary="Test long summary", duration=10.0)
+                )
+            if hasattr(mock_summary, "_on_short_summary"):
+                await mock_summary._on_short_summary(
+                    FinalShortSummary(short_summary="Test short summary", duration=10.0)
+                )
+
+        mock_summary.flush = flush_with_callback
+
+        def init_with_callback(transcript=None, callback=None, on_short_summary=None):
+            mock_summary._callback = callback
+            mock_summary._on_short_summary = on_short_summary
+            return mock_summary
+
+        mock_summary_class.side_effect = init_with_callback
+        yield mock_summary
+
+
+@pytest.mark.asyncio
+async def test_pipeline_main_file_process(
+    tmpdir,
+    mock_transcript_in_db,
+    dummy_file_transcript,
+    dummy_file_diarization,
+    mock_storage,
+    mock_audio_file_writer,
+    mock_waveform_processor,
+    mock_topic_detector,
+    mock_title_processor,
+    mock_summary_processor,
+):
+    """
+    Test the complete PipelineMainFile processing pipeline.
+
+    This test verifies:
+    1. Audio extraction and writing
+    2. Audio upload to storage
+    3. Parallel processing of transcription, diarization, and waveform
+    4. Assembly of transcript with diarization
+    5. Topic detection
+    6. Title and summary generation
+    """
+    # Create a test audio file
+    test_audio_path = Path(__file__).parent / "records" / "test_mathieu_hello.wav"
+
+    # Copy test audio to the transcript's data path as if it was uploaded
+    upload_path = mock_transcript_in_db.data_path / "upload.wav"
+    upload_path.write_bytes(test_audio_path.read_bytes())
+
+    # Also create the audio.mp3 file that would be created by AudioFileWriterProcessor
+    # Since we're mocking AudioFileWriterProcessor, we need to create this manually
+    mp3_path = mock_transcript_in_db.data_path / "audio.mp3"
+    mp3_path.write_bytes(b"mock_mp3_data")
+
+    # Track callback invocations
+    callback_marks = {
+        "on_status": [],
+        "on_duration": [],
+        "on_waveform": [],
+        "on_topic": [],
+        "on_title": [],
+        "on_long_summary": [],
+        "on_short_summary": [],
+    }
+
+    # Create pipeline with mocked callbacks
+    pipeline = PipelineMainFile(transcript_id=mock_transcript_in_db.id)
+
+    # Override callbacks to track invocations
+    async def track_callback(name, data):
+        callback_marks[name].append(data)
+        # Call the original callback
+        original = getattr(PipelineMainFile, name)
+        return await original(pipeline, data)
+
+    for callback_name in callback_marks.keys():
+        setattr(
+            pipeline,
+            callback_name,
+            lambda data, n=callback_name: track_callback(n, data),
+        )
+
+    # Mock av.open for audio processing
+    with patch("reflector.pipelines.main_file_pipeline.av.open") as mock_av:
+        # Mock container for checking video streams
+        mock_container = MagicMock()
+        mock_container.streams.video = []  # No video streams (audio only)
+        mock_container.close = MagicMock()
+
+        # Mock container for decoding audio frames
+        mock_decode_container = MagicMock()
+        mock_decode_container.decode.return_value = iter(
+            [MagicMock()]
+        )  # One mock audio frame
+        mock_decode_container.close = MagicMock()
+
+        # Return different containers for different calls
+        mock_av.side_effect = [mock_container, mock_decode_container]
+
+        # Run the pipeline
+        await pipeline.process(upload_path)
+
+    # Verify audio extraction and writing
+    assert mock_audio_file_writer.push.called
+    assert mock_audio_file_writer.flush.called
+
+    # Verify storage upload
+    assert mock_storage._put_file.called
+    assert mock_storage._get_file_url.called
+
+    # Verify waveform generation
+    assert mock_waveform_processor.flush.called
+    assert mock_waveform_processor.set_pipeline.called
+
+    # Verify topic detection
+    assert mock_topic_detector.push.called
+    assert mock_topic_detector.flush_called
+
+    # Verify title generation
+    assert mock_title_processor.push.called
+    assert mock_title_processor.flush_called
+
+    # Verify summary generation
+    assert mock_summary_processor.push.called
+    assert mock_summary_processor.flush_called
+
+    # Verify callbacks were invoked
+    assert len(callback_marks["on_topic"]) > 0, "Topic callback should be invoked"
+    assert len(callback_marks["on_title"]) > 0, "Title callback should be invoked"
+    assert (
+        len(callback_marks["on_long_summary"]) > 0
+    ), "Long summary callback should be invoked"
+    assert (
+        len(callback_marks["on_short_summary"]) > 0
+    ), "Short summary callback should be invoked"
+
+    print(f"Callback marks: {callback_marks}")
+
+    # Verify the pipeline completed successfully
+    assert pipeline.logger is not None
+    print("PipelineMainFile test completed successfully!")
+
+
+@pytest.mark.asyncio
+async def test_pipeline_main_file_with_video(
+    tmpdir,
+    mock_transcript_in_db,
+    dummy_file_transcript,
+    dummy_file_diarization,
+    mock_storage,
+    mock_audio_file_writer,
+    mock_waveform_processor,
+    mock_topic_detector,
+    mock_title_processor,
+    mock_summary_processor,
+):
+    """
+    Test PipelineMainFile with video input (verifies audio extraction).
+    """
+    # Create a test audio file
+    test_audio_path = Path(__file__).parent / "records" / "test_mathieu_hello.wav"
+
+    # Copy test audio to the transcript's data path as if it was a video upload
+    upload_path = mock_transcript_in_db.data_path / "upload.mp4"
+    upload_path.write_bytes(test_audio_path.read_bytes())
+
+    # Also create the audio.mp3 file that would be created by AudioFileWriterProcessor
+    mp3_path = mock_transcript_in_db.data_path / "audio.mp3"
+    mp3_path.write_bytes(b"mock_mp3_data")
+
+    # Create pipeline
+    pipeline = PipelineMainFile(transcript_id=mock_transcript_in_db.id)
+
+    # Mock av.open for video processing
+    with patch("reflector.pipelines.main_file_pipeline.av.open") as mock_av:
+        # Mock container for checking video streams
+        mock_container = MagicMock()
+        mock_container.streams.video = [MagicMock()]  # Has video streams
+        mock_container.close = MagicMock()
+
+        # Mock container for decoding audio frames
+        mock_decode_container = MagicMock()
+        mock_decode_container.decode.return_value = iter(
+            [MagicMock()]
+        )  # One mock audio frame
+        mock_decode_container.close = MagicMock()
+
+        # Return different containers for different calls
+        mock_av.side_effect = [mock_container, mock_decode_container]
+
+        # Run the pipeline
+        await pipeline.process(upload_path)
+
+    # Verify audio extraction from video
+    assert mock_audio_file_writer.push.called
+    assert mock_audio_file_writer.flush.called
+
+    # Verify the rest of the pipeline completed
+    assert mock_storage._put_file.called
+    assert mock_waveform_processor.flush.called
+    assert mock_topic_detector.push.called
+    assert mock_title_processor.push.called
+    assert mock_summary_processor.push.called
+
+    print("PipelineMainFile video test completed successfully!")
+
+
+@pytest.mark.asyncio
+async def test_pipeline_main_file_no_diarization(
+    tmpdir,
+    mock_transcript_in_db,
+    dummy_file_transcript,
+    mock_storage,
+    mock_audio_file_writer,
+    mock_waveform_processor,
+    mock_topic_detector,
+    mock_title_processor,
+    mock_summary_processor,
+):
+    """
+    Test PipelineMainFile with diarization disabled.
+    """
+    from reflector.settings import settings
+
+    # Disable diarization
+    with patch.object(settings, "DIARIZATION_BACKEND", None):
+        # Create a test audio file
+        test_audio_path = Path(__file__).parent / "records" / "test_mathieu_hello.wav"
+
+        # Copy test audio to the transcript's data path
+        upload_path = mock_transcript_in_db.data_path / "upload.wav"
+        upload_path.write_bytes(test_audio_path.read_bytes())
+
+        # Also create the audio.mp3 file
+        mp3_path = mock_transcript_in_db.data_path / "audio.mp3"
+        mp3_path.write_bytes(b"mock_mp3_data")
+
+        # Create pipeline
+        pipeline = PipelineMainFile(transcript_id=mock_transcript_in_db.id)
+
+        # Mock av.open for audio processing
+        with patch("reflector.pipelines.main_file_pipeline.av.open") as mock_av:
+            # Mock container for checking video streams
+            mock_container = MagicMock()
+            mock_container.streams.video = []  # No video streams
+            mock_container.close = MagicMock()
+
+            # Mock container for decoding audio frames
+            mock_decode_container = MagicMock()
+            mock_decode_container.decode.return_value = iter([MagicMock()])
+            mock_decode_container.close = MagicMock()
+
+            # Return different containers for different calls
+            mock_av.side_effect = [mock_container, mock_decode_container]
+
+            # Run the pipeline
+            await pipeline.process(upload_path)
+
+        # Verify the pipeline completed without diarization
+        assert mock_storage._put_file.called
+        assert mock_waveform_processor.flush.called
+        assert mock_topic_detector.push.called
+        assert mock_title_processor.push.called
+        assert mock_summary_processor.push.called
+
+        print("PipelineMainFile no-diarization test completed successfully!")
+
+
+@pytest.mark.asyncio
+async def test_task_pipeline_file_process(
+    tmpdir,
+    mock_transcript_in_db,
+    dummy_file_transcript,
+    dummy_file_diarization,
+    mock_storage,
+    mock_audio_file_writer,
+    mock_waveform_processor,
+    mock_topic_detector,
+    mock_title_processor,
+    mock_summary_processor,
+):
+    """
+    Test the Celery task entry point for file pipeline processing.
+    """
+    # Direct import of the underlying async function, bypassing the asynctask decorator
+
+    # Create a test audio file in the transcript's data path
+    test_audio_path = Path(__file__).parent / "records" / "test_mathieu_hello.wav"
+    upload_path = mock_transcript_in_db.data_path / "upload.wav"
+    upload_path.write_bytes(test_audio_path.read_bytes())
+
+    # Also create the audio.mp3 file
+    mp3_path = mock_transcript_in_db.data_path / "audio.mp3"
+    mp3_path.write_bytes(b"mock_mp3_data")
+
+    # Mock av.open for audio processing
+    with patch("reflector.pipelines.main_file_pipeline.av.open") as mock_av:
+        # Mock container for checking video streams
+        mock_container = MagicMock()
+        mock_container.streams.video = []  # No video streams
+        mock_container.close = MagicMock()
+
+        # Mock container for decoding audio frames
+        mock_decode_container = MagicMock()
+        mock_decode_container.decode.return_value = iter([MagicMock()])
+        mock_decode_container.close = MagicMock()
+
+        # Return different containers for different calls
+        mock_av.side_effect = [mock_container, mock_decode_container]
+
+        # Get the original async function without the asynctask decorator
+        # The function is wrapped, so we need to call it differently
+        # For now, we test the pipeline directly since the task is just a thin wrapper
+        from reflector.pipelines.main_file_pipeline import PipelineMainFile
+
+        pipeline = PipelineMainFile(transcript_id=mock_transcript_in_db.id)
+        await pipeline.process(upload_path)
+
+    # Verify the pipeline was executed through the task
+    assert mock_audio_file_writer.push.called
+    assert mock_audio_file_writer.flush.called
+    assert mock_storage._put_file.called
+    assert mock_waveform_processor.flush.called
+    assert mock_topic_detector.push.called
+    assert mock_title_processor.push.called
+    assert mock_summary_processor.push.called
+
+    print("task_pipeline_file_process test completed successfully!")
+
+
+@pytest.mark.asyncio
+async def test_pipeline_file_process_no_transcript():
+    """
+    Test the pipeline with a non-existent transcript.
+    """
+    from reflector.pipelines.main_file_pipeline import PipelineMainFile
+
+    # Mock the controller to return None (transcript not found)
+    with patch(
+        "reflector.pipelines.main_file_pipeline.transcripts_controller.get_by_id"
+    ) as mock_get:
+        mock_get.return_value = None
+
+        pipeline = PipelineMainFile(transcript_id=str(uuid4()))
+
+        # Should raise an exception for missing transcript when get_transcript is called
+        with pytest.raises(Exception, match="Transcript not found"):
+            await pipeline.get_transcript()
+
+
+@pytest.mark.asyncio
+async def test_pipeline_file_process_no_audio_file(
+    mock_transcript_in_db,
+):
+    """
+    Test the pipeline when no audio file is found.
+    """
+    from reflector.pipelines.main_file_pipeline import PipelineMainFile
+
+    # Don't create any audio files in the data path
+    # The pipeline's process should handle missing files gracefully
+
+    pipeline = PipelineMainFile(transcript_id=mock_transcript_in_db.id)
+
+    # Try to process a non-existent file
+    non_existent_path = mock_transcript_in_db.data_path / "nonexistent.wav"
+
+    # This should fail when trying to open the file with av
+    with pytest.raises(Exception):
+        await pipeline.process(non_existent_path)

server/tests/test_processors_modal.py

@@ -0,0 +1,265 @@
+"""
+Tests for Modal-based processors using pytest-recording for HTTP recording/playbook
+
+Note: theses tests require full modal configuration to be able to record
+      vcr cassettes
+
+Configuration required for the first recording:
+- TRANSCRIPT_BACKEND=modal
+- TRANSCRIPT_URL=https://xxxxx--reflector-transcriber-parakeet-web.modal.run
+- TRANSCRIPT_MODAL_API_KEY=xxxxx
+- DIARIZATION_BACKEND=modal
+- DIARIZATION_URL=https://xxxxx--reflector-diarizer-web.modal.run
+- DIARIZATION_MODAL_API_KEY=xxxxx
+"""
+
+from unittest.mock import patch
+
+import pytest
+
+from reflector.processors.file_diarization import FileDiarizationInput
+from reflector.processors.file_diarization_modal import FileDiarizationModalProcessor
+from reflector.processors.file_transcript import FileTranscriptInput
+from reflector.processors.file_transcript_modal import FileTranscriptModalProcessor
+from reflector.processors.transcript_diarization_assembler import (
+    TranscriptDiarizationAssemblerInput,
+    TranscriptDiarizationAssemblerProcessor,
+)
+from reflector.processors.types import DiarizationSegment, Transcript, Word
+
+# Public test audio file hosted on S3 specifically for reflector pytests
+TEST_AUDIO_URL = (
+    "https://reflector-github-pytest.s3.us-east-1.amazonaws.com/test_mathieu_hello.mp3"
+)
+
+
+@pytest.mark.asyncio
+async def test_file_transcript_modal_processor_missing_url():
+    with patch("reflector.processors.file_transcript_modal.settings") as mock_settings:
+        mock_settings.TRANSCRIPT_URL = None
+        with pytest.raises(Exception, match="TRANSCRIPT_URL required"):
+            FileTranscriptModalProcessor(modal_api_key="test-api-key")
+
+
+@pytest.mark.asyncio
+async def test_file_diarization_modal_processor_missing_url():
+    with patch("reflector.processors.file_diarization_modal.settings") as mock_settings:
+        mock_settings.DIARIZATION_URL = None
+        with pytest.raises(Exception, match="DIARIZATION_URL required"):
+            FileDiarizationModalProcessor(modal_api_key="test-api-key")
+
+
+@pytest.mark.vcr()
+@pytest.mark.asyncio
+async def test_file_diarization_modal_processor(vcr):
+    """Test FileDiarizationModalProcessor using public audio URL and Modal API"""
+    from reflector.settings import settings
+
+    processor = FileDiarizationModalProcessor(
+        modal_api_key=settings.DIARIZATION_MODAL_API_KEY
+    )
+
+    test_input = FileDiarizationInput(audio_url=TEST_AUDIO_URL)
+    result = await processor._diarize(test_input)
+
+    # Verify the result structure
+    assert result is not None
+    assert hasattr(result, "diarization")
+    assert isinstance(result.diarization, list)
+
+    # Check structure of each diarization segment
+    for segment in result.diarization:
+        assert "start" in segment
+        assert "end" in segment
+        assert "speaker" in segment
+        assert isinstance(segment["start"], (int, float))
+        assert isinstance(segment["end"], (int, float))
+        assert isinstance(segment["speaker"], int)
+        # Basic sanity check - start should be before end
+        assert segment["start"] < segment["end"]
+
+
+@pytest.mark.vcr()
+@pytest.mark.asyncio
+async def test_file_transcript_modal_processor():
+    """Test FileTranscriptModalProcessor using public audio URL and Modal API"""
+    from reflector.settings import settings
+
+    processor = FileTranscriptModalProcessor(
+        modal_api_key=settings.TRANSCRIPT_MODAL_API_KEY
+    )
+
+    test_input = FileTranscriptInput(
+        audio_url=TEST_AUDIO_URL,
+        language="en",
+    )
+
+    # This will record the HTTP interaction on first run, replay on subsequent runs
+    result = await processor._transcript(test_input)
+
+    # Verify the result structure
+    assert result is not None
+    assert hasattr(result, "words")
+    assert isinstance(result.words, list)
+
+    # Check structure of each word if present
+    for word in result.words:
+        assert hasattr(word, "text")
+        assert hasattr(word, "start")
+        assert hasattr(word, "end")
+        assert isinstance(word.start, (int, float))
+        assert isinstance(word.end, (int, float))
+        assert isinstance(word.text, str)
+        # Basic sanity check - start should be before or equal to end
+        assert word.start <= word.end
+
+
+@pytest.mark.asyncio
+async def test_transcript_diarization_assembler_processor():
+    """Test TranscriptDiarizationAssemblerProcessor without VCR (no HTTP requests)"""
+    # Create test transcript with words
+    words = [
+        Word(text="Hello", start=0.0, end=1.0, speaker=0),
+        Word(text=" ", start=1.0, end=1.1, speaker=0),
+        Word(text="world", start=1.1, end=2.0, speaker=0),
+        Word(text=".", start=2.0, end=2.1, speaker=0),
+        Word(text=" ", start=2.1, end=2.2, speaker=0),
+        Word(text="How", start=2.2, end=2.8, speaker=0),
+        Word(text=" ", start=2.8, end=2.9, speaker=0),
+        Word(text="are", start=2.9, end=3.2, speaker=0),
+        Word(text=" ", start=3.2, end=3.3, speaker=0),
+        Word(text="you", start=3.3, end=3.8, speaker=0),
+        Word(text="?", start=3.8, end=3.9, speaker=0),
+    ]
+    transcript = Transcript(words=words)
+
+    # Create test diarization segments
+    diarization = [
+        DiarizationSegment(start=0.0, end=2.1, speaker=0),
+        DiarizationSegment(start=2.1, end=3.9, speaker=1),
+    ]
+
+    # Create processor and test input
+    processor = TranscriptDiarizationAssemblerProcessor()
+    test_input = TranscriptDiarizationAssemblerInput(
+        transcript=transcript, diarization=diarization
+    )
+
+    # Track emitted results
+    emitted_results = []
+
+    async def capture_result(result):
+        emitted_results.append(result)
+
+    processor.on(capture_result)
+
+    # Process the input
+    await processor.push(test_input)
+
+    # Verify result was emitted
+    assert len(emitted_results) == 1
+    result = emitted_results[0]
+
+    # Verify result structure
+    assert isinstance(result, Transcript)
+    assert len(result.words) == len(words)
+
+    # Verify speaker assignments were applied
+    # Words 0-3 (indices) should be speaker 0 (time 0.0-2.0)
+    # Words 4-10 (indices) should be speaker 1 (time 2.1-3.9)
+    for i in range(4):  # First 4 words (Hello, space, world, .)
+        assert (
+            result.words[i].speaker == 0
+        ), f"Word {i} '{result.words[i].text}' should be speaker 0, got {result.words[i].speaker}"
+
+    for i in range(4, 11):  # Remaining words (space, How, space, are, space, you, ?)
+        assert (
+            result.words[i].speaker == 1
+        ), f"Word {i} '{result.words[i].text}' should be speaker 1, got {result.words[i].speaker}"
+
+
+@pytest.mark.asyncio
+async def test_transcript_diarization_assembler_no_diarization():
+    """Test TranscriptDiarizationAssemblerProcessor with no diarization data"""
+    # Create test transcript
+    words = [Word(text="Hello", start=0.0, end=1.0, speaker=0)]
+    transcript = Transcript(words=words)
+
+    # Create processor and test input with empty diarization
+    processor = TranscriptDiarizationAssemblerProcessor()
+    test_input = TranscriptDiarizationAssemblerInput(
+        transcript=transcript, diarization=[]
+    )
+
+    # Track emitted results
+    emitted_results = []
+
+    async def capture_result(result):
+        emitted_results.append(result)
+
+    processor.on(capture_result)
+
+    # Process the input
+    await processor.push(test_input)
+
+    # Verify original transcript was returned unchanged
+    assert len(emitted_results) == 1
+    result = emitted_results[0]
+    assert result is transcript  # Should be the same object
+    assert result.words[0].speaker == 0  # Original speaker unchanged
+
+
+@pytest.mark.vcr()
+@pytest.mark.asyncio
+async def test_full_modal_pipeline_integration(vcr):
+    """Integration test: Transcription -> Diarization -> Assembly
+
+    This test demonstrates the full pipeline:
+    1. Run transcription via Modal
+    2. Run diarization via Modal
+    3. Assemble transcript with diarization
+    """
+    from reflector.settings import settings
+
+    # Step 1: Transcription
+    transcript_processor = FileTranscriptModalProcessor(
+        modal_api_key=settings.TRANSCRIPT_MODAL_API_KEY
+    )
+    transcript_input = FileTranscriptInput(audio_url=TEST_AUDIO_URL, language="en")
+    transcript = await transcript_processor._transcript(transcript_input)
+
+    # Step 2: Diarization
+    diarization_processor = FileDiarizationModalProcessor(
+        modal_api_key=settings.DIARIZATION_MODAL_API_KEY
+    )
+    diarization_input = FileDiarizationInput(audio_url=TEST_AUDIO_URL)
+    diarization_result = await diarization_processor._diarize(diarization_input)
+
+    # Step 3: Assembly
+    assembler = TranscriptDiarizationAssemblerProcessor()
+    assembly_input = TranscriptDiarizationAssemblerInput(
+        transcript=transcript, diarization=diarization_result.diarization
+    )
+
+    # Track assembled result
+    assembled_results = []
+
+    async def capture_result(result):
+        assembled_results.append(result)
+
+    assembler.on(capture_result)
+
+    await assembler.push(assembly_input)
+
+    # Verify the full pipeline worked
+    assert len(assembled_results) == 1
+    final_transcript = assembled_results[0]
+
+    # Verify the final transcript has the original words with updated speaker info
+    assert isinstance(final_transcript, Transcript)
+    assert len(final_transcript.words) == len(transcript.words)
+    assert len(final_transcript.words) > 0
+
+    # Verify some words have been assigned speakers from diarization
+    speakers_found = set(word.speaker for word in final_transcript.words)
+    assert len(speakers_found) > 0  # At least some speaker assignments

server/tests/test_processors_pipeline.py

@@ -2,10 +2,13 @@ import pytest
 
 
 @pytest.mark.asyncio
+@pytest.mark.parametrize("enable_diarization", [False, True])
 async def test_basic_process(
     dummy_transcript,
     dummy_llm,
     dummy_processors,
+    enable_diarization,
+    dummy_diarization,
 ):
     # goal is to start the server, and send rtc audio to it
     # validate the events received
@@ -28,12 +31,31 @@ async def test_basic_process(
 
     # invoke the process and capture events
     path = Path(__file__).parent / "records" / "test_mathieu_hello.wav"
-    await process_audio_file(path.as_posix(), event_callback)
-    print(marks)
+
+    if enable_diarization:
+        # Test with diarization - may fail if pyannote.audio is not installed
+        try:
+            await process_audio_file(
+                path.as_posix(), event_callback, enable_diarization=True
+            )
+        except SystemExit:
+            pytest.skip("pyannote.audio not installed - skipping diarization test")
+    else:
+        # Test without diarization - should always work
+        await process_audio_file(
+            path.as_posix(), event_callback, enable_diarization=False
+        )
+
+    print(f"Diarization: {enable_diarization}, Marks: {marks}")
 
     # validate the events
-    assert marks["TranscriptLinerProcessor"] == 1
-    assert marks["TranscriptTranslatorPassthroughProcessor"] == 1
+    # Each processor should be called for each audio segment processed
+    # The final processors (Topic, Title, Summary) should be called once at the end
+    assert marks["TranscriptLinerProcessor"] > 0
+    assert marks["TranscriptTranslatorPassthroughProcessor"] > 0
     assert marks["TranscriptTopicDetectorProcessor"] == 1
     assert marks["TranscriptFinalSummaryProcessor"] == 1
     assert marks["TranscriptFinalTitleProcessor"] == 1
+
+    if enable_diarization:
+        assert marks["TestAudioDiarizationProcessor"] == 1

server/tests/test_search.py

@@ -0,0 +1,485 @@
+"""Tests for full-text search functionality."""
+
+import json
+from datetime import datetime, timezone
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from reflector.db import get_database
+from reflector.db.search import (
+    SearchController,
+    SearchParameters,
+    SearchResult,
+    search_controller,
+)
+from reflector.db.transcripts import SourceKind, transcripts
+
+
+@pytest.mark.asyncio
+async def test_search_postgresql_only():
+    params = SearchParameters(query_text="any query here")
+    results, total = await search_controller.search_transcripts(params)
+    assert results == []
+    assert total == 0
+
+    params_empty = SearchParameters(query_text="")
+    results_empty, total_empty = await search_controller.search_transcripts(
+        params_empty
+    )
+    assert isinstance(results_empty, list)
+    assert isinstance(total_empty, int)
+
+
+@pytest.mark.asyncio
+async def test_search_with_empty_query():
+    """Test that empty query returns all transcripts."""
+    params = SearchParameters(query_text="")
+    results, total = await search_controller.search_transcripts(params)
+
+    assert isinstance(results, list)
+    assert isinstance(total, int)
+    if len(results) > 1:
+        for i in range(len(results) - 1):
+            assert results[i].created_at >= results[i + 1].created_at
+
+
+@pytest.mark.asyncio
+async def test_empty_transcript_title_only_match():
+    """Test that transcripts with title-only matches return empty snippets."""
+    test_id = "test-empty-9b3f2a8d"
+
+    try:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+
+        test_data = {
+            "id": test_id,
+            "name": "Empty Transcript",
+            "title": "Empty Meeting",
+            "status": "completed",
+            "locked": False,
+            "duration": 0.0,
+            "created_at": datetime.now(timezone.utc),
+            "short_summary": None,
+            "long_summary": None,
+            "topics": json.dumps([]),
+            "events": json.dumps([]),
+            "participants": json.dumps([]),
+            "source_language": "en",
+            "target_language": "en",
+            "reviewed": False,
+            "audio_location": "local",
+            "share_mode": "private",
+            "source_kind": "room",
+            "webvtt": None,
+            "user_id": "test-user-1",
+        }
+
+        await get_database().execute(transcripts.insert().values(**test_data))
+
+        params = SearchParameters(query_text="empty", user_id="test-user-1")
+        results, total = await search_controller.search_transcripts(params)
+
+        assert total >= 1
+        found = next((r for r in results if r.id == test_id), None)
+        assert found is not None, "Should find transcript by title match"
+        assert found.search_snippets == []
+        assert found.total_match_count == 0
+
+    finally:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+        await get_database().disconnect()
+
+
+@pytest.mark.asyncio
+async def test_search_with_long_summary():
+    """Test that long_summary content is searchable."""
+    test_id = "test-long-summary-8a9f3c2d"
+
+    try:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+
+        test_data = {
+            "id": test_id,
+            "name": "Test Long Summary",
+            "title": "Regular Meeting",
+            "status": "completed",
+            "locked": False,
+            "duration": 1800.0,
+            "created_at": datetime.now(timezone.utc),
+            "short_summary": "Brief overview",
+            "long_summary": "Detailed discussion about quantum computing applications and blockchain technology integration",
+            "topics": json.dumps([]),
+            "events": json.dumps([]),
+            "participants": json.dumps([]),
+            "source_language": "en",
+            "target_language": "en",
+            "reviewed": False,
+            "audio_location": "local",
+            "share_mode": "private",
+            "source_kind": "room",
+            "webvtt": """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+Basic meeting content without special keywords.""",
+            "user_id": "test-user-2",
+        }
+
+        await get_database().execute(transcripts.insert().values(**test_data))
+
+        params = SearchParameters(query_text="quantum computing", user_id="test-user-2")
+        results, total = await search_controller.search_transcripts(params)
+
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find transcript by long_summary content"
+
+        test_result = next((r for r in results if r.id == test_id), None)
+        assert test_result
+        assert len(test_result.search_snippets) > 0
+        assert "quantum computing" in test_result.search_snippets[0].lower()
+
+    finally:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+        await get_database().disconnect()
+
+
+@pytest.mark.asyncio
+async def test_postgresql_search_with_data():
+    test_id = "test-search-e2e-7f3a9b2c"
+
+    try:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+
+        test_data = {
+            "id": test_id,
+            "name": "Test Search Transcript",
+            "title": "Engineering Planning Meeting Q4 2024",
+            "status": "completed",
+            "locked": False,
+            "duration": 1800.0,
+            "created_at": datetime.now(timezone.utc),
+            "short_summary": "Team discussed search implementation",
+            "long_summary": "The engineering team met to plan the search feature",
+            "topics": json.dumps([]),
+            "events": json.dumps([]),
+            "participants": json.dumps([]),
+            "source_language": "en",
+            "target_language": "en",
+            "reviewed": False,
+            "audio_location": "local",
+            "share_mode": "private",
+            "source_kind": "room",
+            "webvtt": """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+Welcome to our engineering planning meeting for Q4 2024.
+
+00:00:10.000 --> 00:00:20.000
+Today we'll discuss the implementation of full-text search.
+
+00:00:20.000 --> 00:00:30.000
+The search feature should support complex queries with ranking.
+
+00:00:30.000 --> 00:00:40.000
+We need to implement PostgreSQL tsvector for better performance.""",
+            "user_id": "test-user-3",
+        }
+
+        await get_database().execute(transcripts.insert().values(**test_data))
+
+        params = SearchParameters(query_text="planning", user_id="test-user-3")
+        results, total = await search_controller.search_transcripts(params)
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find test transcript by title word"
+
+        params = SearchParameters(query_text="tsvector", user_id="test-user-3")
+        results, total = await search_controller.search_transcripts(params)
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find test transcript by webvtt content"
+
+        params = SearchParameters(
+            query_text="engineering planning", user_id="test-user-3"
+        )
+        results, total = await search_controller.search_transcripts(params)
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find test transcript by multiple words"
+
+        test_result = next((r for r in results if r.id == test_id), None)
+        if test_result:
+            assert test_result.title == "Engineering Planning Meeting Q4 2024"
+            assert test_result.status == "completed"
+            assert test_result.duration == 1800.0
+            assert 0 <= test_result.rank <= 1, "Rank should be normalized to 0-1"
+
+        params = SearchParameters(
+            query_text="tsvector OR nosuchword", user_id="test-user-3"
+        )
+        results, total = await search_controller.search_transcripts(params)
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find test transcript with OR query"
+
+        params = SearchParameters(
+            query_text='"full-text search"', user_id="test-user-3"
+        )
+        results, total = await search_controller.search_transcripts(params)
+        assert total >= 1
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find test transcript by exact phrase"
+
+    finally:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+        await get_database().disconnect()
+
+
+@pytest.fixture
+def sample_search_params():
+    """Create sample search parameters for testing."""
+    return SearchParameters(
+        query_text="test query",
+        limit=20,
+        offset=0,
+        user_id="test-user",
+        room_id="room1",
+    )
+
+
+@pytest.fixture
+def mock_db_result():
+    """Create a mock database result."""
+    return {
+        "id": "test-transcript-id",
+        "title": "Test Transcript",
+        "created_at": datetime(2024, 6, 15, tzinfo=timezone.utc),
+        "duration": 3600.0,
+        "status": "completed",
+        "user_id": "test-user",
+        "room_id": "room1",
+        "source_kind": SourceKind.LIVE,
+        "webvtt": "WEBVTT\n\n00:00:00.000 --> 00:00:05.000\nThis is a test transcript",
+        "rank": 0.95,
+    }
+
+
+class TestSearchParameters:
+    """Test SearchParameters model validation and functionality."""
+
+    def test_search_parameters_with_available_filters(self):
+        """Test creating SearchParameters with currently available filter options."""
+        params = SearchParameters(
+            query_text="search term",
+            limit=50,
+            offset=10,
+            user_id="user123",
+            room_id="room1",
+        )
+
+        assert params.query_text == "search term"
+        assert params.limit == 50
+        assert params.offset == 10
+        assert params.user_id == "user123"
+        assert params.room_id == "room1"
+
+    def test_search_parameters_defaults(self):
+        """Test SearchParameters with default values."""
+        params = SearchParameters(query_text="test")
+
+        assert params.query_text == "test"
+        assert params.limit == 20
+        assert params.offset == 0
+        assert params.user_id is None
+        assert params.room_id is None
+
+
+class TestSearchControllerFilters:
+    """Test SearchController functionality with various filters."""
+
+    @pytest.mark.asyncio
+    async def test_search_with_source_kind_filter(self):
+        """Test search filtering by source_kind."""
+        controller = SearchController()
+        with (
+            patch("reflector.db.search.is_postgresql", return_value=True),
+            patch("reflector.db.search.get_database") as mock_db,
+        ):
+            mock_db.return_value.fetch_all = AsyncMock(return_value=[])
+            mock_db.return_value.fetch_val = AsyncMock(return_value=0)
+
+            params = SearchParameters(query_text="test", source_kind=SourceKind.LIVE)
+
+            results, total = await controller.search_transcripts(params)
+
+            assert results == []
+            assert total == 0
+
+            mock_db.return_value.fetch_all.assert_called_once()
+
+    @pytest.mark.asyncio
+    async def test_search_with_single_room_id(self):
+        """Test search filtering by single room ID (currently supported)."""
+        controller = SearchController()
+        with (
+            patch("reflector.db.search.is_postgresql", return_value=True),
+            patch("reflector.db.search.get_database") as mock_db,
+        ):
+            mock_db.return_value.fetch_all = AsyncMock(return_value=[])
+            mock_db.return_value.fetch_val = AsyncMock(return_value=0)
+
+            params = SearchParameters(
+                query_text="test",
+                room_id="room1",
+            )
+
+            results, total = await controller.search_transcripts(params)
+
+            assert results == []
+            assert total == 0
+            mock_db.return_value.fetch_all.assert_called_once()
+
+    @pytest.mark.asyncio
+    async def test_search_result_includes_available_fields(self, mock_db_result):
+        """Test that search results include available fields like source_kind."""
+        controller = SearchController()
+        with (
+            patch("reflector.db.search.is_postgresql", return_value=True),
+            patch("reflector.db.search.get_database") as mock_db,
+        ):
+
+            class MockRow:
+                def __init__(self, data):
+                    self._data = data
+                    self._mapping = data
+
+                def __iter__(self):
+                    return iter(self._data.items())
+
+                def __getitem__(self, key):
+                    return self._data[key]
+
+                def keys(self):
+                    return self._data.keys()
+
+            mock_row = MockRow(mock_db_result)
+
+            mock_db.return_value.fetch_all = AsyncMock(return_value=[mock_row])
+            mock_db.return_value.fetch_val = AsyncMock(return_value=1)
+
+            params = SearchParameters(query_text="test")
+
+            results, total = await controller.search_transcripts(params)
+
+            assert total == 1
+            assert len(results) == 1
+
+            result = results[0]
+            assert isinstance(result, SearchResult)
+            assert result.id == "test-transcript-id"
+            assert result.title == "Test Transcript"
+            assert result.rank == 0.95
+
+
+class TestSearchEndpointParsing:
+    """Test parameter parsing in the search endpoint."""
+
+    def test_parse_comma_separated_room_ids(self):
+        """Test parsing comma-separated room IDs."""
+        room_ids_str = "room1,room2,room3"
+        parsed = [rid.strip() for rid in room_ids_str.split(",") if rid.strip()]
+        assert parsed == ["room1", "room2", "room3"]
+
+        room_ids_str = "room1, room2 , room3"
+        parsed = [rid.strip() for rid in room_ids_str.split(",") if rid.strip()]
+        assert parsed == ["room1", "room2", "room3"]
+
+        room_ids_str = "room1,,room3,"
+        parsed = [rid.strip() for rid in room_ids_str.split(",") if rid.strip()]
+        assert parsed == ["room1", "room3"]
+
+    def test_parse_source_kind(self):
+        """Test parsing source_kind values."""
+        for kind_str in ["live", "file", "room"]:
+            parsed = SourceKind(kind_str)
+            assert parsed == SourceKind(kind_str)
+
+        with pytest.raises(ValueError):
+            SourceKind("invalid_kind")
+
+
+class TestSearchResultModel:
+    """Test SearchResult model and serialization."""
+
+    def test_search_result_with_available_fields(self):
+        """Test SearchResult model with currently available fields populated."""
+        result = SearchResult(
+            id="test-id",
+            title="Test Title",
+            user_id="user-123",
+            room_id="room-456",
+            source_kind=SourceKind.ROOM,
+            created_at=datetime(2024, 6, 15, tzinfo=timezone.utc),
+            status="completed",
+            rank=0.85,
+            duration=1800.5,
+            search_snippets=["snippet 1", "snippet 2"],
+        )
+
+        assert result.id == "test-id"
+        assert result.title == "Test Title"
+        assert result.user_id == "user-123"
+        assert result.room_id == "room-456"
+        assert result.status == "completed"
+        assert result.rank == 0.85
+        assert result.duration == 1800.5
+        assert len(result.search_snippets) == 2
+
+    def test_search_result_with_optional_fields_none(self):
+        """Test SearchResult model with optional fields as None."""
+        result = SearchResult(
+            id="test-id",
+            source_kind=SourceKind.FILE,
+            created_at=datetime.now(timezone.utc),
+            status="processing",
+            rank=0.5,
+            search_snippets=[],
+            title=None,
+            user_id=None,
+            room_id=None,
+            duration=None,
+        )
+
+        assert result.title is None
+        assert result.user_id is None
+        assert result.room_id is None
+        assert result.duration is None
+
+    def test_search_result_datetime_field(self):
+        """Test that SearchResult accepts datetime field."""
+        result = SearchResult(
+            id="test-id",
+            source_kind=SourceKind.LIVE,
+            created_at=datetime(2024, 6, 15, 12, 30, 45, tzinfo=timezone.utc),
+            status="completed",
+            rank=0.9,
+            duration=None,
+            search_snippets=[],
+        )
+
+        assert result.created_at == datetime(
+            2024, 6, 15, 12, 30, 45, tzinfo=timezone.utc
+        )

server/tests/test_search_long_summary.py

@@ -0,0 +1,166 @@
+"""Tests for long_summary in search functionality."""
+
+import json
+from datetime import datetime, timezone
+
+import pytest
+
+from reflector.db import get_database
+from reflector.db.search import SearchParameters, search_controller
+from reflector.db.transcripts import transcripts
+
+
+@pytest.mark.asyncio
+async def test_long_summary_snippet_prioritization():
+    """Test that snippets from long_summary are prioritized over webvtt content."""
+    test_id = "test-snippet-priority-3f9a2b8c"
+
+    try:
+        # Clean up any existing test data
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+
+        test_data = {
+            "id": test_id,
+            "name": "Test Snippet Priority",
+            "title": "Meeting About Projects",
+            "status": "completed",
+            "locked": False,
+            "duration": 1800.0,
+            "created_at": datetime.now(timezone.utc),
+            "short_summary": "Project discussion",
+            "long_summary": (
+                "The team discussed advanced robotics applications including "
+                "autonomous navigation systems and sensor fusion techniques. "
+                "Robotics development will focus on real-time processing."
+            ),
+            "topics": json.dumps([]),
+            "events": json.dumps([]),
+            "participants": json.dumps([]),
+            "source_language": "en",
+            "target_language": "en",
+            "reviewed": False,
+            "audio_location": "local",
+            "share_mode": "private",
+            "source_kind": "room",
+            "webvtt": """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+We talked about many different topics today.
+
+00:00:10.000 --> 00:00:20.000
+The robotics project is making good progress.
+
+00:00:20.000 --> 00:00:30.000
+We need to consider various implementation approaches.""",
+            "user_id": "test-user-priority",
+        }
+
+        await get_database().execute(transcripts.insert().values(**test_data))
+
+        # Search for "robotics" which appears in both long_summary and webvtt
+        params = SearchParameters(query_text="robotics", user_id="test-user-priority")
+        results, total = await search_controller.search_transcripts(params)
+
+        assert total >= 1
+        test_result = next((r for r in results if r.id == test_id), None)
+        assert test_result, "Should find the test transcript"
+
+        snippets = test_result.search_snippets
+        assert len(snippets) > 0, "Should have at least one snippet"
+
+        # The first snippets should be from long_summary (more detailed content)
+        first_snippet = snippets[0].lower()
+        assert (
+            "advanced robotics" in first_snippet or "autonomous" in first_snippet
+        ), f"First snippet should be from long_summary with detailed content. Got: {snippets[0]}"
+
+        # With max 3 snippets, we should get both from long_summary and webvtt
+        assert len(snippets) <= 3, "Should respect max snippets limit"
+
+        # All snippets should contain the search term
+        for snippet in snippets:
+            assert (
+                "robotics" in snippet.lower()
+            ), f"Snippet should contain search term: {snippet}"
+
+    finally:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+        await get_database().disconnect()
+
+
+@pytest.mark.asyncio
+async def test_long_summary_only_search():
+    """Test searching for content that only exists in long_summary."""
+    test_id = "test-long-only-8b3c9f2a"
+
+    try:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+
+        test_data = {
+            "id": test_id,
+            "name": "Test Long Only",
+            "title": "Standard Meeting",
+            "status": "completed",
+            "locked": False,
+            "duration": 1800.0,
+            "created_at": datetime.now(timezone.utc),
+            "short_summary": "Team sync",
+            "long_summary": (
+                "Detailed analysis of cryptocurrency market trends and "
+                "decentralized finance protocols. Discussion included "
+                "yield farming strategies and liquidity pool mechanics."
+            ),
+            "topics": json.dumps([]),
+            "events": json.dumps([]),
+            "participants": json.dumps([]),
+            "source_language": "en",
+            "target_language": "en",
+            "reviewed": False,
+            "audio_location": "local",
+            "share_mode": "private",
+            "source_kind": "room",
+            "webvtt": """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+Team meeting about general project updates.
+
+00:00:10.000 --> 00:00:20.000
+Discussion of timeline and deliverables.""",
+            "user_id": "test-user-long",
+        }
+
+        await get_database().execute(transcripts.insert().values(**test_data))
+
+        # Search for terms only in long_summary
+        params = SearchParameters(query_text="cryptocurrency", user_id="test-user-long")
+        results, total = await search_controller.search_transcripts(params)
+
+        found = any(r.id == test_id for r in results)
+        assert found, "Should find transcript by long_summary-only content"
+
+        test_result = next((r for r in results if r.id == test_id), None)
+        assert test_result
+        assert len(test_result.search_snippets) > 0
+
+        # Verify the snippet is about cryptocurrency
+        snippet = test_result.search_snippets[0].lower()
+        assert "cryptocurrency" in snippet, "Snippet should contain the search term"
+
+        # Search for "yield farming" - a more specific term
+        params2 = SearchParameters(query_text="yield farming", user_id="test-user-long")
+        results2, total2 = await search_controller.search_transcripts(params2)
+
+        found2 = any(r.id == test_id for r in results2)
+        assert found2, "Should find transcript by specific long_summary phrase"
+
+    finally:
+        await get_database().execute(
+            transcripts.delete().where(transcripts.c.id == test_id)
+        )
+        await get_database().disconnect()

server/tests/test_search_snippets.py

@@ -0,0 +1,534 @@
+"""Unit tests for search snippet generation."""
+
+from reflector.db.search import (
+    SnippetCandidate,
+    SnippetGenerator,
+    WebVTTProcessor,
+)
+
+
+class TestExtractWebVTT:
+    """Test WebVTT text extraction."""
+
+    def test_extract_webvtt_with_speakers(self):
+        """Test extraction removes speaker tags and timestamps."""
+        webvtt = """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+<v Speaker0>Hello world, this is a test.
+
+00:00:10.000 --> 00:00:20.000
+<v Speaker1>Indeed it is a test of WebVTT parsing.
+"""
+        result = WebVTTProcessor.extract_text(webvtt)
+        assert "Hello world, this is a test" in result
+        assert "Indeed it is a test" in result
+        assert "<v Speaker" not in result
+        assert "00:00" not in result
+        assert "-->" not in result
+
+    def test_extract_empty_webvtt(self):
+        """Test empty WebVTT returns empty string."""
+        assert WebVTTProcessor.extract_text("") == ""
+
+    def test_extract_malformed_webvtt(self):
+        """Test malformed WebVTT returns empty string."""
+        result = WebVTTProcessor.extract_text("Not a valid WebVTT")
+        assert result == ""
+
+
+class TestGenerateSnippets:
+    """Test snippet generation from plain text."""
+
+    def test_multiple_matches(self):
+        """Test finding multiple occurrences of search term in long text."""
+        separator = " This is filler text. " * 20
+        text = (
+            "Python is great for machine learning."
+            + separator
+            + "Many companies use Python for data science."
+            + separator
+            + "Python has excellent libraries for analysis."
+            + separator
+            + "The Python community is very supportive."
+        )
+
+        snippets = SnippetGenerator.generate(text, "Python")
+        assert len(snippets) >= 2
+
+        for snippet in snippets:
+            assert "python" in snippet.lower()
+
+    def test_single_match(self):
+        """Test single occurrence returns one snippet."""
+        text = "This document discusses artificial intelligence and its applications."
+        snippets = SnippetGenerator.generate(text, "artificial intelligence")
+
+        assert len(snippets) == 1
+        assert "artificial intelligence" in snippets[0].lower()
+
+    def test_no_matches(self):
+        """Test no matches returns empty list."""
+        text = "This is some random text without the search term."
+        snippets = SnippetGenerator.generate(text, "machine learning")
+
+        assert snippets == []
+
+    def test_case_insensitive_search(self):
+        """Test search is case insensitive."""
+        text = (
+            "MACHINE LEARNING is important for modern applications. "
+            + "It requires lots of data and computational resources. " * 5
+            + "Machine Learning rocks and transforms industries. "
+            + "Deep learning is a subset of it. " * 5
+            + "Finally, machine learning will shape our future."
+        )
+
+        snippets = SnippetGenerator.generate(text, "machine learning")
+
+        assert len(snippets) >= 2
+        for snippet in snippets:
+            assert "machine learning" in snippet.lower()
+
+    def test_partial_match_fallback(self):
+        """Test fallback to first word when exact phrase not found."""
+        text = "We use machine intelligence for processing."
+        snippets = SnippetGenerator.generate(text, "machine learning")
+
+        assert len(snippets) == 1
+        assert "machine" in snippets[0].lower()
+
+    def test_snippet_ellipsis(self):
+        """Test ellipsis added for truncated snippets."""
+        text = "a " * 100 + "TARGET_WORD special content here" + " b" * 100
+        snippets = SnippetGenerator.generate(text, "TARGET_WORD")
+
+        assert len(snippets) == 1
+        assert "..." in snippets[0]
+        assert "TARGET_WORD" in snippets[0]
+
+    def test_overlapping_snippets_deduplicated(self):
+        """Test overlapping matches don't create duplicate snippets."""
+        text = "test test test word" * 10
+        snippets = SnippetGenerator.generate(text, "test")
+
+        assert len(snippets) <= 3
+        assert len(snippets) == len(set(snippets))
+
+    def test_empty_inputs(self):
+        """Test empty text or search term returns empty list."""
+        assert SnippetGenerator.generate("", "search") == []
+        assert SnippetGenerator.generate("text", "") == []
+        assert SnippetGenerator.generate("", "") == []
+
+    def test_max_snippets_limit(self):
+        """Test respects max_snippets parameter."""
+        separator = " filler " * 50
+        text = ("Python is amazing" + separator) * 10
+
+        snippets_1 = SnippetGenerator.generate(text, "Python", max_snippets=1)
+        assert len(snippets_1) == 1
+
+        snippets_2 = SnippetGenerator.generate(text, "Python", max_snippets=2)
+        assert len(snippets_2) == 2
+
+        snippets_5 = SnippetGenerator.generate(text, "Python", max_snippets=5)
+        assert len(snippets_5) == 5
+
+    def test_snippet_length(self):
+        """Test snippet length is reasonable."""
+        text = "word " * 200
+        snippets = SnippetGenerator.generate(text, "word")
+
+        for snippet in snippets:
+            assert len(snippet) <= 200
+
+
+class TestFullPipeline:
+    """Test the complete WebVTT to snippets pipeline."""
+
+    def test_webvtt_to_snippets_integration(self):
+        """Test full pipeline from WebVTT to search snippets."""
+        webvtt = (
+            """WEBVTT
+
+00:00:00.000 --> 00:00:10.000
+<v Speaker0>Let's discuss machine learning applications in modern technology.
+
+00:00:10.000 --> 00:00:20.000
+<v Speaker1>"""
+            + "Various industries are adopting new technologies. " * 10
+            + """
+
+00:00:20.000 --> 00:00:30.000
+<v Speaker2>Machine learning is revolutionizing healthcare and diagnostics.
+
+00:00:30.000 --> 00:00:40.000
+<v Speaker3>"""
+            + "Financial markets show interesting patterns. " * 10
+            + """
+
+00:00:40.000 --> 00:00:50.000
+<v Speaker0>Machine learning in education provides personalized experiences.
+"""
+        )
+
+        plain_text = WebVTTProcessor.extract_text(webvtt)
+        snippets = SnippetGenerator.generate(plain_text, "machine learning")
+
+        assert len(snippets) >= 1
+        assert len(snippets) <= 3
+
+        for snippet in snippets:
+            assert "machine learning" in snippet.lower()
+            assert "<v Speaker" not in snippet
+            assert "00:00" not in snippet
+            assert "-->" not in snippet
+
+
+class TestMultiWordQueryBehavior:
+    """Tests for multi-word query behavior and exact phrase matching."""
+
+    def test_multi_word_query_snippet_behavior(self):
+        """Test that multi-word queries generate snippets based on exact phrase matching."""
+        sample_text = """This is a sample transcript where user Alice is talking.
+        Later in the conversation, jordan mentions something important.
+        The user jordan collaboration was successful.
+        Another user named Bob joins the discussion."""
+
+        user_snippets = SnippetGenerator.generate(sample_text, "user")
+        assert len(user_snippets) == 2, "Should find 2 snippets for 'user'"
+
+        jordan_snippets = SnippetGenerator.generate(sample_text, "jordan")
+        assert len(jordan_snippets) >= 1, "Should find at least 1 snippet for 'jordan'"
+
+        multi_word_snippets = SnippetGenerator.generate(sample_text, "user jordan")
+        assert len(multi_word_snippets) == 1, (
+            "Should return exactly 1 snippet for 'user jordan' "
+            "(only the exact phrase match, not individual word occurrences)"
+        )
+
+        snippet = multi_word_snippets[0]
+        assert (
+            "user jordan" in snippet.lower()
+        ), "The snippet should contain the exact phrase 'user jordan'"
+
+        assert (
+            "alice" not in snippet.lower()
+        ), "The snippet should not include the first standalone 'user' with Alice"
+
+    def test_multi_word_query_without_exact_match(self):
+        """Test snippet generation when exact phrase is not found."""
+        sample_text = """User Alice is here. Bob and jordan are talking.
+        Later jordan mentions something. The user is happy."""
+
+        snippets = SnippetGenerator.generate(sample_text, "user jordan")
+
+        assert (
+            len(snippets) >= 1
+        ), "Should find at least 1 snippet when falling back to first word"
+
+        all_snippets_text = " ".join(snippets).lower()
+        assert (
+            "user" in all_snippets_text
+        ), "Snippets should contain 'user' (the first word)"
+
+    def test_exact_phrase_at_text_boundaries(self):
+        """Test snippet generation when exact phrase appears at text boundaries."""
+
+        text_start = "user jordan started the meeting. Other content here."
+        snippets = SnippetGenerator.generate(text_start, "user jordan")
+        assert len(snippets) == 1
+        assert "user jordan" in snippets[0].lower()
+
+        text_end = "Other content here. The meeting ended with user jordan"
+        snippets = SnippetGenerator.generate(text_end, "user jordan")
+        assert len(snippets) == 1
+        assert "user jordan" in snippets[0].lower()
+
+    def test_multi_word_query_matches_words_appearing_separately_and_together(self):
+        """Test that multi-word queries prioritize exact phrase matches over individual word occurrences."""
+        sample_text = """This is a sample transcript where user Alice is talking.
+        Later in the conversation, jordan mentions something important.
+        The user jordan collaboration was successful.
+        Another user named Bob joins the discussion."""
+
+        search_query = "user jordan"
+        snippets = SnippetGenerator.generate(sample_text, search_query)
+
+        assert len(snippets) == 1, (
+            f"Expected exactly 1 snippet for '{search_query}' when exact phrase exists, "
+            f"got {len(snippets)}. Should ignore individual word occurrences."
+        )
+
+        snippet = snippets[0]
+
+        assert (
+            search_query in snippet.lower()
+        ), f"Snippet should contain the exact phrase '{search_query}'. Got: {snippet}"
+
+        assert (
+            "jordan mentions" in snippet.lower()
+        ), f"Snippet should include context before the exact phrase match. Got: {snippet}"
+
+        assert (
+            "alice" not in snippet.lower()
+        ), f"Snippet should not include separate occurrences of individual words. Got: {snippet}"
+
+        text_2 = """The alpha version was released.
+        Beta testing started yesterday.
+        The alpha beta integration is complete."""
+
+        snippets_2 = SnippetGenerator.generate(text_2, "alpha beta")
+        assert len(snippets_2) == 1, "Should return 1 snippet for exact phrase match"
+        assert "alpha beta" in snippets_2[0].lower(), "Should contain exact phrase"
+        assert (
+            "version" not in snippets_2[0].lower()
+        ), "Should not include first separate occurrence"
+
+
+class TestSnippetGenerationEnhanced:
+    """Additional snippet generation tests from test_search_enhancements.py."""
+
+    def test_snippet_generation_from_webvtt(self):
+        """Test snippet generation from WebVTT content."""
+        webvtt_content = """WEBVTT
+
+00:00:00.000 --> 00:00:05.000
+This is the beginning of the transcript
+
+00:00:05.000 --> 00:00:10.000
+The search term appears here in the middle
+
+00:00:10.000 --> 00:00:15.000
+And this is the end of the content"""
+
+        plain_text = WebVTTProcessor.extract_text(webvtt_content)
+        snippets = SnippetGenerator.generate(plain_text, "search term")
+
+        assert len(snippets) > 0
+        assert any("search term" in snippet.lower() for snippet in snippets)
+
+    def test_extract_webvtt_text_with_malformed_variations(self):
+        """Test WebVTT extraction with various malformed content."""
+        malformed_vtt = "This is not valid WebVTT content"
+        result = WebVTTProcessor.extract_text(malformed_vtt)
+        assert result == ""
+
+        partial_vtt = "WEBVTT\nNo timestamps here"
+        result = WebVTTProcessor.extract_text(partial_vtt)
+        assert result == "" or "No timestamps" not in result
+
+
+class TestPureFunctions:
+    """Test the pure functions extracted for functional programming."""
+
+    def test_find_all_matches(self):
+        """Test finding all match positions in text."""
+        text = "Python is great. Python is powerful. I love Python."
+        matches = list(SnippetGenerator.find_all_matches(text, "Python"))
+        assert matches == [0, 17, 44]
+
+        matches = list(SnippetGenerator.find_all_matches(text, "python"))
+        assert matches == [0, 17, 44]
+
+        matches = list(SnippetGenerator.find_all_matches(text, "Ruby"))
+        assert matches == []
+
+        matches = list(SnippetGenerator.find_all_matches("", "test"))
+        assert matches == []
+        matches = list(SnippetGenerator.find_all_matches("test", ""))
+        assert matches == []
+
+    def test_create_snippet(self):
+        """Test creating a snippet from a match position."""
+        text = "This is a long text with the word Python in the middle and more text after."
+
+        snippet = SnippetGenerator.create_snippet(text, 35, max_length=150)
+        assert "Python" in snippet.text()
+        assert snippet.start >= 0
+        assert snippet.end <= len(text)
+        assert isinstance(snippet, SnippetCandidate)
+
+        assert len(snippet.text()) > 0
+        assert snippet.start <= snippet.end
+
+        long_text = "A" * 200
+        snippet = SnippetGenerator.create_snippet(long_text, 100, max_length=50)
+        assert snippet.text().startswith("...")
+        assert snippet.text().endswith("...")
+
+        snippet = SnippetGenerator.create_snippet("short text", 0, max_length=100)
+        assert snippet.start == 0
+        assert "short text" in snippet.text()
+
+    def test_filter_non_overlapping(self):
+        """Test filtering overlapping snippets."""
+        candidates = [
+            SnippetCandidate(_text="First snippet", start=0, _original_text_length=100),
+            SnippetCandidate(_text="Overlapping", start=10, _original_text_length=100),
+            SnippetCandidate(
+                _text="Third snippet", start=40, _original_text_length=100
+            ),
+            SnippetCandidate(
+                _text="Fourth snippet", start=65, _original_text_length=100
+            ),
+        ]
+
+        filtered = list(SnippetGenerator.filter_non_overlapping(iter(candidates)))
+        assert filtered == [
+            "First snippet...",
+            "...Third snippet...",
+            "...Fourth snippet...",
+        ]
+
+        filtered = list(SnippetGenerator.filter_non_overlapping(iter([])))
+        assert filtered == []
+
+    def test_generate_integration(self):
+        """Test the main SnippetGenerator.generate function."""
+        text = "Machine learning is amazing. Machine learning transforms data. Learn machine learning today."
+
+        snippets = SnippetGenerator.generate(text, "machine learning")
+        assert len(snippets) <= 3
+        assert all("machine learning" in s.lower() for s in snippets)
+
+        snippets = SnippetGenerator.generate(text, "machine learning", max_snippets=2)
+        assert len(snippets) <= 2
+
+        snippets = SnippetGenerator.generate(text, "machine vision")
+        assert len(snippets) > 0
+        assert any("machine" in s.lower() for s in snippets)
+
+    def test_extract_webvtt_text_basic(self):
+        """Test WebVTT text extraction (basic test, full tests exist elsewhere)."""
+        webvtt = """WEBVTT
+
+00:00:00.000 --> 00:00:02.000
+Hello world
+
+00:00:02.000 --> 00:00:04.000
+This is a test"""
+
+        result = WebVTTProcessor.extract_text(webvtt)
+        assert "Hello world" in result
+        assert "This is a test" in result
+
+        # Test empty input
+        assert WebVTTProcessor.extract_text("") == ""
+        assert WebVTTProcessor.extract_text(None) == ""
+
+    def test_generate_webvtt_snippets(self):
+        """Test generating snippets from WebVTT content."""
+        webvtt = """WEBVTT
+
+00:00:00.000 --> 00:00:02.000
+Python programming is great
+
+00:00:02.000 --> 00:00:04.000
+Learn Python today"""
+
+        snippets = WebVTTProcessor.generate_snippets(webvtt, "Python")
+        assert len(snippets) > 0
+        assert any("Python" in s for s in snippets)
+
+        snippets = WebVTTProcessor.generate_snippets("", "Python")
+        assert snippets == []
+
+    def test_from_summary(self):
+        """Test generating snippets from summary text."""
+        summary = "This meeting discussed Python development and machine learning applications."
+
+        snippets = SnippetGenerator.from_summary(summary, "Python")
+        assert len(snippets) > 0
+        assert any("Python" in s for s in snippets)
+
+        long_summary = "Python " * 20
+        snippets = SnippetGenerator.from_summary(long_summary, "Python")
+        assert len(snippets) <= 2
+
+    def test_combine_sources(self):
+        """Test combining snippets from multiple sources."""
+        summary = "Python is a great programming language."
+        webvtt = """WEBVTT
+
+00:00:00.000 --> 00:00:02.000
+Learn Python programming
+
+00:00:02.000 --> 00:00:04.000
+Python is powerful"""
+
+        snippets, total_count = SnippetGenerator.combine_sources(
+            summary, webvtt, "Python", max_total=3
+        )
+        assert len(snippets) <= 3
+        assert len(snippets) > 0
+        assert total_count > 0
+
+        snippets, total_count = SnippetGenerator.combine_sources(
+            summary, None, "Python", max_total=3
+        )
+        assert len(snippets) > 0
+        assert all("Python" in s for s in snippets)
+        assert total_count == 1
+
+        snippets, total_count = SnippetGenerator.combine_sources(
+            None, webvtt, "Python", max_total=3
+        )
+        assert len(snippets) > 0
+        assert total_count == 2
+
+        long_summary = "Python " * 10
+        snippets, total_count = SnippetGenerator.combine_sources(
+            long_summary, webvtt, "Python", max_total=2
+        )
+        assert len(snippets) == 2
+        assert total_count >= 10
+
+    def test_match_counting_sum_logic(self):
+        """Test that match counting correctly sums matches from both sources."""
+        summary = "data science uses data analysis and data mining techniques"
+        webvtt = """WEBVTT
+
+00:00:00.000 --> 00:00:02.000
+Big data processing
+
+00:00:02.000 --> 00:00:04.000
+data visualization and data storage"""
+
+        snippets, total_count = SnippetGenerator.combine_sources(
+            summary, webvtt, "data", max_total=3
+        )
+        assert total_count == 6
+        assert len(snippets) <= 3
+
+        summary_snippets, summary_count = SnippetGenerator.combine_sources(
+            summary, None, "data", max_total=3
+        )
+        assert summary_count == 3
+
+        webvtt_snippets, webvtt_count = SnippetGenerator.combine_sources(
+            None, webvtt, "data", max_total=3
+        )
+        assert webvtt_count == 3
+
+        snippets_empty, count_empty = SnippetGenerator.combine_sources(
+            None, None, "data", max_total=3
+        )
+        assert snippets_empty == []
+        assert count_empty == 0
+
+    def test_edge_cases(self):
+        """Test edge cases for the pure functions."""
+        text = "Test with special: @#$%^&*() characters"
+        snippets = SnippetGenerator.generate(text, "@#$%")
+        assert len(snippets) > 0
+
+        long_query = "a" * 100
+        snippets = SnippetGenerator.generate("Some text", long_query)
+        assert snippets == []
+
+        text = "Unicode test: café, naïve, 日本語"
+        snippets = SnippetGenerator.generate(text, "café")
+        assert len(snippets) > 0
+        assert "café" in snippets[0]

server/tests/test_transcripts.py

@@ -1,15 +1,11 @@
 from contextlib import asynccontextmanager
 
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.asyncio
-async def test_transcript_create():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_create(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["name"] == "test"
     assert response.json()["status"] == "idle"
@@ -23,71 +19,62 @@ async def test_transcript_create():
 
 
 @pytest.mark.asyncio
-async def test_transcript_get_update_name():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_get_update_name(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["name"] == "test"
 
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test"
 
-        response = await ac.patch(f"/transcripts/{tid}", json={"name": "test2"})
+    response = await client.patch(f"/transcripts/{tid}", json={"name": "test2"})
     assert response.status_code == 200
     assert response.json()["name"] == "test2"
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test2"
 
 
 @pytest.mark.asyncio
-async def test_transcript_get_update_locked():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_get_update_locked(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["locked"] is False
 
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["locked"] is False
 
-        response = await ac.patch(f"/transcripts/{tid}", json={"locked": True})
+    response = await client.patch(f"/transcripts/{tid}", json={"locked": True})
     assert response.status_code == 200
     assert response.json()["locked"] is True
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["locked"] is True
 
 
 @pytest.mark.asyncio
-async def test_transcript_get_update_summary():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_get_update_summary(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["long_summary"] is None
     assert response.json()["short_summary"] is None
 
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["long_summary"] is None
     assert response.json()["short_summary"] is None
 
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{tid}",
         json={"long_summary": "test_long", "short_summary": "test_short"},
     )
@@ -95,52 +82,46 @@ async def test_transcript_get_update_summary():
     assert response.json()["long_summary"] == "test_long"
     assert response.json()["short_summary"] == "test_short"
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["long_summary"] == "test_long"
     assert response.json()["short_summary"] == "test_short"
 
 
 @pytest.mark.asyncio
-async def test_transcript_get_update_title():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_get_update_title(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["title"] is None
 
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["title"] is None
 
-        response = await ac.patch(f"/transcripts/{tid}", json={"title": "test_title"})
+    response = await client.patch(f"/transcripts/{tid}", json={"title": "test_title"})
     assert response.status_code == 200
     assert response.json()["title"] == "test_title"
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["title"] == "test_title"
 
 
 @pytest.mark.asyncio
-async def test_transcripts_list_anonymous():
+async def test_transcripts_list_anonymous(client):
     # XXX this test is a bit fragile, as it depends on the storage which
     #     is shared between tests
-    from reflector.app import app
     from reflector.settings import settings
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.get("/transcripts")
+    response = await client.get("/transcripts")
     assert response.status_code == 401
 
     # if public mode, it should be allowed
     try:
         settings.PUBLIC_MODE = True
-        async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-            response = await ac.get("/transcripts")
+        response = await client.get("/transcripts")
         assert response.status_code == 200
     finally:
         settings.PUBLIC_MODE = False
@@ -197,21 +178,19 @@ async def authenticated_client2():
 
 
 @pytest.mark.asyncio
-async def test_transcripts_list_authenticated(authenticated_client):
+async def test_transcripts_list_authenticated(authenticated_client, client):
     # XXX this test is a bit fragile, as it depends on the storage which
     #     is shared between tests
-    from reflector.app import app
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "testxx1"})
+    response = await client.post("/transcripts", json={"name": "testxx1"})
     assert response.status_code == 200
     assert response.json()["name"] == "testxx1"
 
-        response = await ac.post("/transcripts", json={"name": "testxx2"})
+    response = await client.post("/transcripts", json={"name": "testxx2"})
     assert response.status_code == 200
     assert response.json()["name"] == "testxx2"
 
-        response = await ac.get("/transcripts")
+    response = await client.get("/transcripts")
     assert response.status_code == 200
     assert len(response.json()["items"]) >= 2
     names = [t["name"] for t in response.json()["items"]]
@@ -220,44 +199,38 @@ async def test_transcripts_list_authenticated(authenticated_client):
 
 
 @pytest.mark.asyncio
-async def test_transcript_delete():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "testdel1"})
+async def test_transcript_delete(client):
+    response = await client.post("/transcripts", json={"name": "testdel1"})
     assert response.status_code == 200
     assert response.json()["name"] == "testdel1"
 
     tid = response.json()["id"]
-        response = await ac.delete(f"/transcripts/{tid}")
+    response = await client.delete(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["status"] == "ok"
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 404
 
 
 @pytest.mark.asyncio
-async def test_transcript_mark_reviewed():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_mark_reviewed(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["name"] == "test"
     assert response.json()["reviewed"] is False
 
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test"
     assert response.json()["reviewed"] is False
 
-        response = await ac.patch(f"/transcripts/{tid}", json={"reviewed": True})
+    response = await client.patch(f"/transcripts/{tid}", json={"reviewed": True})
     assert response.status_code == 200
     assert response.json()["reviewed"] is True
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["reviewed"] is True

server/tests/test_transcripts_audio_download.py

@@ -2,20 +2,17 @@ import shutil
 from pathlib import Path
 
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.fixture
-async def fake_transcript(tmpdir):
-    from reflector.app import app
+async def fake_transcript(tmpdir, client):
     from reflector.settings import settings
     from reflector.views.transcripts import transcripts_controller
 
     settings.DATA_DIR = Path(tmpdir)
 
     # create a transcript
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.post("/transcripts", json={"name": "Test audio download"})
+    response = await client.post("/transcripts", json={"name": "Test audio download"})
     assert response.status_code == 200
     tid = response.json()["id"]
 
@@ -39,17 +36,17 @@ async def fake_transcript(tmpdir):
         ["/mp3", "audio/mpeg"],
     ],
 )
-async def test_transcript_audio_download(fake_transcript, url_suffix, content_type):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.get(f"/transcripts/{fake_transcript.id}/audio{url_suffix}")
+async def test_transcript_audio_download(
+    fake_transcript, url_suffix, content_type, client
+):
+    response = await client.get(f"/transcripts/{fake_transcript.id}/audio{url_suffix}")
     assert response.status_code == 200
     assert response.headers["content-type"] == content_type
 
     # test get 404
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.get(f"/transcripts/{fake_transcript.id}XXX/audio{url_suffix}")
+    response = await client.get(
+        f"/transcripts/{fake_transcript.id}XXX/audio{url_suffix}"
+    )
     assert response.status_code == 404
 
 
@@ -61,18 +58,16 @@ async def test_transcript_audio_download(fake_transcript, url_suffix, content_ty
     ],
 )
 async def test_transcript_audio_download_head(
-    fake_transcript, url_suffix, content_type
+    fake_transcript, url_suffix, content_type, client
 ):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.head(f"/transcripts/{fake_transcript.id}/audio{url_suffix}")
+    response = await client.head(f"/transcripts/{fake_transcript.id}/audio{url_suffix}")
     assert response.status_code == 200
     assert response.headers["content-type"] == content_type
 
     # test head 404
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.head(f"/transcripts/{fake_transcript.id}XXX/audio{url_suffix}")
+    response = await client.head(
+        f"/transcripts/{fake_transcript.id}XXX/audio{url_suffix}"
+    )
     assert response.status_code == 404
 
 
@@ -84,12 +79,9 @@ async def test_transcript_audio_download_head(
     ],
 )
 async def test_transcript_audio_download_range(
-    fake_transcript, url_suffix, content_type
+    fake_transcript, url_suffix, content_type, client
 ):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.get(
+    response = await client.get(
         f"/transcripts/{fake_transcript.id}/audio{url_suffix}",
         headers={"range": "bytes=0-100"},
     )
@@ -107,12 +99,9 @@ async def test_transcript_audio_download_range(
     ],
 )
 async def test_transcript_audio_download_range_with_seek(
-    fake_transcript, url_suffix, content_type
+    fake_transcript, url_suffix, content_type, client
 ):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.get(
+    response = await client.get(
         f"/transcripts/{fake_transcript.id}/audio{url_suffix}",
         headers={"range": "bytes=100-"},
     )
@@ -122,13 +111,10 @@ async def test_transcript_audio_download_range_with_seek(
 
 
 @pytest.mark.asyncio
-async def test_transcript_delete_with_audio(fake_transcript):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-    response = await ac.delete(f"/transcripts/{fake_transcript.id}")
+async def test_transcript_delete_with_audio(fake_transcript, client):
+    response = await client.delete(f"/transcripts/{fake_transcript.id}")
     assert response.status_code == 200
     assert response.json()["status"] == "ok"
 
-    response = await ac.get(f"/transcripts/{fake_transcript.id}")
+    response = await client.get(f"/transcripts/{fake_transcript.id}")
     assert response.status_code == 404

server/tests/test_transcripts_participants.py

@@ -1,19 +1,15 @@
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.asyncio
-async def test_transcript_participants():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_participants(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["participants"] == []
 
     # create a participant
     transcript_id = response.json()["id"]
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants", json={"name": "test"}
     )
     assert response.status_code == 200
@@ -22,7 +18,7 @@ async def test_transcript_participants():
     assert response.json()["name"] == "test"
 
     # create another one with a speaker
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test2", "speaker": 1},
     )
@@ -32,28 +28,25 @@ async def test_transcript_participants():
     assert response.json()["name"] == "test2"
 
     # get all participants via transcript
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
     assert len(response.json()["participants"]) == 2
 
     # get participants via participants endpoint
-        response = await ac.get(f"/transcripts/{transcript_id}/participants")
+    response = await client.get(f"/transcripts/{transcript_id}/participants")
     assert response.status_code == 200
     assert len(response.json()) == 2
 
 
 @pytest.mark.asyncio
-async def test_transcript_participants_same_speaker():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_participants_same_speaker(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["participants"] == []
     transcript_id = response.json()["id"]
 
     # create a participant
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test", "speaker": 1},
     )
@@ -61,7 +54,7 @@ async def test_transcript_participants_same_speaker():
     assert response.json()["speaker"] == 1
 
     # create another one with the same speaker
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test2", "speaker": 1},
     )
@@ -69,17 +62,14 @@ async def test_transcript_participants_same_speaker():
 
 
 @pytest.mark.asyncio
-async def test_transcript_participants_update_name():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_participants_update_name(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["participants"] == []
     transcript_id = response.json()["id"]
 
     # create a participant
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test", "speaker": 1},
     )
@@ -88,7 +78,7 @@ async def test_transcript_participants_update_name():
 
     # update the participant
     participant_id = response.json()["id"]
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/participants/{participant_id}",
         json={"name": "test2"},
     )
@@ -96,31 +86,28 @@ async def test_transcript_participants_update_name():
     assert response.json()["name"] == "test2"
 
     # verify the participant was updated
-        response = await ac.get(
+    response = await client.get(
         f"/transcripts/{transcript_id}/participants/{participant_id}"
     )
     assert response.status_code == 200
     assert response.json()["name"] == "test2"
 
     # verify the participant was updated in transcript
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
     assert len(response.json()["participants"]) == 1
     assert response.json()["participants"][0]["name"] == "test2"
 
 
 @pytest.mark.asyncio
-async def test_transcript_participants_update_speaker():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test"})
+async def test_transcript_participants_update_speaker(client):
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["participants"] == []
     transcript_id = response.json()["id"]
 
     # create a participant
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test", "speaker": 1},
     )
@@ -128,7 +115,7 @@ async def test_transcript_participants_update_speaker():
     participant1_id = response.json()["id"]
 
     # create another participant
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={"name": "test2", "speaker": 2},
     )
@@ -136,27 +123,27 @@ async def test_transcript_participants_update_speaker():
     participant2_id = response.json()["id"]
 
     # update the participant, refused as speaker is already taken
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/participants/{participant2_id}",
         json={"speaker": 1},
     )
     assert response.status_code == 400
 
     # delete the participant 1
-        response = await ac.delete(
+    response = await client.delete(
         f"/transcripts/{transcript_id}/participants/{participant1_id}"
     )
     assert response.status_code == 200
 
     # update the participant 2 again, should be accepted now
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/participants/{participant2_id}",
         json={"speaker": 1},
     )
     assert response.status_code == 200
 
     # ensure participant2 name is still there
-        response = await ac.get(
+    response = await client.get(
         f"/transcripts/{transcript_id}/participants/{participant2_id}"
     )
     assert response.status_code == 200

server/tests/test_transcripts_process.py

@@ -1,7 +1,26 @@
 import asyncio
+import time
 
 import pytest
-from httpx import AsyncClient
+from httpx import ASGITransport, AsyncClient
+
+
+@pytest.fixture
+async def app_lifespan():
+    from asgi_lifespan import LifespanManager
+
+    from reflector.app import app
+
+    async with LifespanManager(app) as manager:
+        yield manager.app
+
+
+@pytest.fixture
+async def client(app_lifespan):
+    yield AsyncClient(
+        transport=ASGITransport(app=app_lifespan),
+        base_url="http://test/v1",
+    )
 
 
 @pytest.mark.usefixtures("setup_database")
@@ -10,23 +29,21 @@ from httpx import AsyncClient
 @pytest.mark.asyncio
 async def test_transcript_process(
     tmpdir,
+    whisper_transcript,
     dummy_llm,
     dummy_processors,
     dummy_diarization,
     dummy_storage,
+    client,
 ):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-
     # create a transcript
-    response = await ac.post("/transcripts", json={"name": "test"})
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["status"] == "idle"
     tid = response.json()["id"]
 
     # upload mp3
-    response = await ac.post(
+    response = await client.post(
         f"/transcripts/{tid}/record/upload?chunk_number=0&total_chunks=1",
         files={
             "chunk": (
@@ -39,30 +56,38 @@ async def test_transcript_process(
     assert response.status_code == 200
     assert response.json()["status"] == "ok"
 
-    # wait for processing to finish
-    while True:
+    # wait for processing to finish (max 10 minutes)
+    timeout_seconds = 600  # 10 minutes
+    start_time = time.monotonic()
+    while (time.monotonic() - start_time) < timeout_seconds:
         # fetch the transcript and check if it is ended
-        resp = await ac.get(f"/transcripts/{tid}")
+        resp = await client.get(f"/transcripts/{tid}")
         assert resp.status_code == 200
         if resp.json()["status"] in ("ended", "error"):
             break
         await asyncio.sleep(1)
+    else:
+        pytest.fail(f"Initial processing timed out after {timeout_seconds} seconds")
 
     # restart the processing
-    response = await ac.post(
+    response = await client.post(
         f"/transcripts/{tid}/process",
     )
     assert response.status_code == 200
     assert response.json()["status"] == "ok"
 
-    # wait for processing to finish
-    while True:
+    # wait for processing to finish (max 10 minutes)
+    timeout_seconds = 600  # 10 minutes
+    start_time = time.monotonic()
+    while (time.monotonic() - start_time) < timeout_seconds:
         # fetch the transcript and check if it is ended
-        resp = await ac.get(f"/transcripts/{tid}")
+        resp = await client.get(f"/transcripts/{tid}")
         assert resp.status_code == 200
         if resp.json()["status"] in ("ended", "error"):
             break
         await asyncio.sleep(1)
+    else:
+        pytest.fail(f"Restart processing timed out after {timeout_seconds} seconds")
 
     # check the transcript is ended
     transcript = resp.json()
@@ -71,7 +96,7 @@ async def test_transcript_process(
     assert transcript["title"] == "Llm Title"
 
     # check topics and transcript
-    response = await ac.get(f"/transcripts/{tid}/topics")
+    response = await client.get(f"/transcripts/{tid}/topics")
     assert response.status_code == 200
     assert len(response.json()) == 1
     assert "want to share" in response.json()[0]["transcript"]

server/tests/test_transcripts_recording_deletion.py

@@ -0,0 +1,34 @@
+from datetime import datetime, timezone
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from reflector.db.recordings import Recording, recordings_controller
+from reflector.db.transcripts import SourceKind, transcripts_controller
+
+
+@pytest.mark.asyncio
+async def test_recording_deleted_with_transcript():
+    recording = await recordings_controller.create(
+        Recording(
+            bucket_name="test-bucket",
+            object_key="recording.mp4",
+            recorded_at=datetime.now(timezone.utc),
+        )
+    )
+    transcript = await transcripts_controller.add(
+        name="Test Transcript",
+        source_kind=SourceKind.ROOM,
+        recording_id=recording.id,
+    )
+
+    with patch("reflector.db.transcripts.get_recordings_storage") as mock_get_storage:
+        storage_instance = mock_get_storage.return_value
+        storage_instance.delete_file = AsyncMock()
+
+        await transcripts_controller.remove_by_id(transcript.id)
+
+        storage_instance.delete_file.assert_awaited_once_with(recording.object_key)
+
+    assert await recordings_controller.get_by_id(recording.id) is None
+    assert await transcripts_controller.get_by_id(transcript.id) is None

server/tests/test_transcripts_rtc_ws.py

@@ -6,10 +6,10 @@
 import asyncio
 import json
 import threading
+import time
 from pathlib import Path
 
 import pytest
-from httpx import AsyncClient
 from httpx_ws import aconnect_ws
 from uvicorn import Config, Server
 
@@ -21,34 +21,97 @@ class ThreadedUvicorn:
 
     async def start(self):
         self.thread.start()
-        while not self.server.started:
+        timeout_seconds = 600  # 10 minutes
+        start_time = time.monotonic()
+        while (
+            not self.server.started
+            and (time.monotonic() - start_time) < timeout_seconds
+        ):
             await asyncio.sleep(0.1)
+        if not self.server.started:
+            raise TimeoutError(
+                f"Server failed to start after {timeout_seconds} seconds"
+            )
 
     def stop(self):
         if self.thread.is_alive():
             self.server.should_exit = True
-            while self.thread.is_alive():
-                continue
+            timeout_seconds = 600  # 10 minutes
+            start_time = time.time()
+            while (
+                self.thread.is_alive() and (time.time() - start_time) < timeout_seconds
+            ):
+                time.sleep(0.1)
+            if self.thread.is_alive():
+                raise TimeoutError(
+                    f"Thread failed to stop after {timeout_seconds} seconds"
+                )
 
 
 @pytest.fixture
-async def appserver(tmpdir, setup_database, celery_session_app, celery_session_worker):
+def appserver(tmpdir, setup_database, celery_session_app, celery_session_worker):
+    import threading
+
     from reflector.app import app
+    from reflector.db import get_database
     from reflector.settings import settings
 
     DATA_DIR = settings.DATA_DIR
     settings.DATA_DIR = Path(tmpdir)
 
-    # start server
+    # start server in a separate thread with its own event loop
     host = "127.0.0.1"
     port = 1255
-    config = Config(app=app, host=host, port=port)
-    server = ThreadedUvicorn(config)
-    await server.start()
+    server_started = threading.Event()
+    server_exception = None
+    server_instance = None
 
-    yield (server, host, port)
+    def run_server():
+        nonlocal server_exception, server_instance
+        try:
+            # Create a new event loop for this thread
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+
+            config = Config(app=app, host=host, port=port, loop=loop)
+            server_instance = Server(config)
+
+            async def start_server():
+                # Initialize database connection in this event loop
+                database = get_database()
+                await database.connect()
+                try:
+                    await server_instance.serve()
+                finally:
+                    await database.disconnect()
+
+            # Signal that server is starting
+            server_started.set()
+            loop.run_until_complete(start_server())
+        except Exception as e:
+            server_exception = e
+            server_started.set()
+        finally:
+            loop.close()
+
+    server_thread = threading.Thread(target=run_server, daemon=True)
+    server_thread.start()
+
+    # Wait for server to start
+    server_started.wait(timeout=30)
+    if server_exception:
+        raise server_exception
+
+    # Wait a bit more for the server to be fully ready
+    time.sleep(1)
+
+    yield server_instance, host, port
+
+    # Stop server
+    if server_instance:
+        server_instance.should_exit = True
+        server_thread.join(timeout=30)
 
-    server.stop()
     settings.DATA_DIR = DATA_DIR
 
 
@@ -71,6 +134,7 @@ async def test_transcript_rtc_and_websocket(
     dummy_storage,
     fake_mp3_upload,
     appserver,
+    client,
 ):
     # goal: start the server, exchange RTC, receive websocket events
     # because of that, we need to start the server in a thread
@@ -79,8 +143,7 @@ async def test_transcript_rtc_and_websocket(
 
     # create a transcript
     base_url = f"http://{host}:{port}/v1"
-    ac = AsyncClient(base_url=base_url)
-    response = await ac.post("/transcripts", json={"name": "Test RTC"})
+    response = await client.post("/transcripts", json={"name": "Test RTC"})
     assert response.status_code == 200
     tid = response.json()["id"]
 
@@ -92,12 +155,16 @@ async def test_transcript_rtc_and_websocket(
         async with aconnect_ws(f"{base_url}/transcripts/{tid}/events") as ws:
             print("Test websocket: CONNECTED")
             try:
-                while True:
+                timeout_seconds = 600  # 10 minutes
+                start_time = time.monotonic()
+                while (time.monotonic() - start_time) < timeout_seconds:
                     msg = await ws.receive_json()
                     print(f"Test websocket: JSON {msg}")
                     if msg is None:
                         break
                     events.append(msg)
+                else:
+                    print(f"Test websocket: TIMEOUT after {timeout_seconds} seconds")
             except Exception as e:
                 print(f"Test websocket: EXCEPTION {e}")
             finally:
@@ -121,11 +188,11 @@ async def test_transcript_rtc_and_websocket(
 
     url = f"{base_url}/transcripts/{tid}/record/webrtc"
     path = Path(__file__).parent / "records" / "test_short.wav"
-    client = StreamClient(signaling, url=url, play_from=path.as_posix())
-    await client.start()
+    stream_client = StreamClient(signaling, url=url, play_from=path.as_posix())
+    await stream_client.start()
 
-    timeout = 20
-    while not client.is_ended():
+    timeout = 120
+    while not stream_client.is_ended():
         await asyncio.sleep(1)
         timeout -= 1
         if timeout < 0:
@@ -133,21 +200,24 @@ async def test_transcript_rtc_and_websocket(
 
     # XXX aiortc is long to close the connection
     # instead of waiting a long time, we just send a STOP
-    client.channel.send(json.dumps({"cmd": "STOP"}))
-    await client.stop()
+    stream_client.channel.send(json.dumps({"cmd": "STOP"}))
+    await stream_client.stop()
 
     # wait the processing to finish
-    timeout = 20
+    timeout = 120
     while True:
         # fetch the transcript and check if it is ended
-        resp = await ac.get(f"/transcripts/{tid}")
+        resp = await client.get(f"/transcripts/{tid}")
         assert resp.status_code == 200
         if resp.json()["status"] in ("ended", "error"):
             break
         await asyncio.sleep(1)
+        timeout -= 1
+        if timeout < 0:
+            raise TimeoutError("Timeout while waiting for transcript to be ended")
 
     if resp.json()["status"] != "ended":
-        raise TimeoutError("Timeout while waiting for transcript to be ended")
+        raise TimeoutError("Transcript processing failed")
 
     # stop websocket task
     websocket_task.cancel()
@@ -190,7 +260,7 @@ async def test_transcript_rtc_and_websocket(
     ev = events[eventnames.index("WAVEFORM")]
     assert isinstance(ev["data"]["waveform"], list)
     assert len(ev["data"]["waveform"]) >= 250
-    waveform_resp = await ac.get(f"/transcripts/{tid}/audio/waveform")
+    waveform_resp = await client.get(f"/transcripts/{tid}/audio/waveform")
     assert waveform_resp.status_code == 200
     assert waveform_resp.headers["content-type"] == "application/json"
     assert isinstance(waveform_resp.json()["data"], list)
@@ -210,7 +280,7 @@ async def test_transcript_rtc_and_websocket(
     assert "DURATION" in eventnames
 
     # check that audio/mp3 is available
-    audio_resp = await ac.get(f"/transcripts/{tid}/audio/mp3")
+    audio_resp = await client.get(f"/transcripts/{tid}/audio/mp3")
     assert audio_resp.status_code == 200
     assert audio_resp.headers["Content-Type"] == "audio/mpeg"
 
@@ -229,6 +299,7 @@ async def test_transcript_rtc_and_websocket_and_fr(
     dummy_storage,
     fake_mp3_upload,
     appserver,
+    client,
 ):
     # goal: start the server, exchange RTC, receive websocket events
     # because of that, we need to start the server in a thread
@@ -238,8 +309,7 @@ async def test_transcript_rtc_and_websocket_and_fr(
 
     # create a transcript
     base_url = f"http://{host}:{port}/v1"
-    ac = AsyncClient(base_url=base_url)
-    response = await ac.post(
+    response = await client.post(
         "/transcripts", json={"name": "Test RTC", "target_language": "fr"}
     )
     assert response.status_code == 200
@@ -253,12 +323,16 @@ async def test_transcript_rtc_and_websocket_and_fr(
         async with aconnect_ws(f"{base_url}/transcripts/{tid}/events") as ws:
             print("Test websocket: CONNECTED")
             try:
-                while True:
+                timeout_seconds = 600  # 10 minutes
+                start_time = time.monotonic()
+                while (time.monotonic() - start_time) < timeout_seconds:
                     msg = await ws.receive_json()
                     print(f"Test websocket: JSON {msg}")
                     if msg is None:
                         break
                     events.append(msg)
+                else:
+                    print(f"Test websocket: TIMEOUT after {timeout_seconds} seconds")
             except Exception as e:
                 print(f"Test websocket: EXCEPTION {e}")
             finally:
@@ -282,11 +356,11 @@ async def test_transcript_rtc_and_websocket_and_fr(
 
     url = f"{base_url}/transcripts/{tid}/record/webrtc"
     path = Path(__file__).parent / "records" / "test_short.wav"
-    client = StreamClient(signaling, url=url, play_from=path.as_posix())
-    await client.start()
+    stream_client = StreamClient(signaling, url=url, play_from=path.as_posix())
+    await stream_client.start()
 
-    timeout = 20
-    while not client.is_ended():
+    timeout = 120
+    while not stream_client.is_ended():
         await asyncio.sleep(1)
         timeout -= 1
         if timeout < 0:
@@ -294,25 +368,28 @@ async def test_transcript_rtc_and_websocket_and_fr(
 
     # XXX aiortc is long to close the connection
     # instead of waiting a long time, we just send a STOP
-    client.channel.send(json.dumps({"cmd": "STOP"}))
+    stream_client.channel.send(json.dumps({"cmd": "STOP"}))
 
     # wait the processing to finish
     await asyncio.sleep(2)
 
-    await client.stop()
+    await stream_client.stop()
 
     # wait the processing to finish
-    timeout = 20
+    timeout = 120
     while True:
         # fetch the transcript and check if it is ended
-        resp = await ac.get(f"/transcripts/{tid}")
+        resp = await client.get(f"/transcripts/{tid}")
         assert resp.status_code == 200
         if resp.json()["status"] == "ended":
             break
         await asyncio.sleep(1)
+        timeout -= 1
+        if timeout < 0:
+            raise TimeoutError("Timeout while waiting for transcript to be ended")
 
     if resp.json()["status"] != "ended":
-        raise TimeoutError("Timeout while waiting for transcript to be ended")
+        raise TimeoutError("Transcript processing failed")
 
     await asyncio.sleep(2)
 

server/tests/test_transcripts_speaker.py

@@ -1,20 +1,16 @@
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.asyncio
-async def test_transcript_reassign_speaker(fake_transcript_with_topics):
-    from reflector.app import app
-
+async def test_transcript_reassign_speaker(fake_transcript_with_topics, client):
     transcript_id = fake_transcript_with_topics.id
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
     # check the transcript exists
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
 
     # check initial topics of the transcript
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -31,7 +27,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert topics[1]["segments"][0]["speaker"] == 0
 
     # reassign speaker
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "speaker": 1,
@@ -42,7 +38,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -59,7 +55,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert topics[1]["segments"][0]["speaker"] == 0
 
     # reassign speaker, middle of 2 topics
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "speaker": 2,
@@ -70,7 +66,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -89,7 +85,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert topics[1]["segments"][1]["speaker"] == 0
 
     # reassign speaker, everything
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "speaker": 4,
@@ -100,7 +96,7 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -118,18 +114,15 @@ async def test_transcript_reassign_speaker(fake_transcript_with_topics):
 
 
 @pytest.mark.asyncio
-async def test_transcript_merge_speaker(fake_transcript_with_topics):
-    from reflector.app import app
-
+async def test_transcript_merge_speaker(fake_transcript_with_topics, client):
     transcript_id = fake_transcript_with_topics.id
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
     # check the transcript exists
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
 
     # check initial topics of the transcript
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -141,7 +134,7 @@ async def test_transcript_merge_speaker(fake_transcript_with_topics):
     assert topics[1]["words"][1]["speaker"] == 0
 
     # reassign speaker
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "speaker": 1,
@@ -152,7 +145,7 @@ async def test_transcript_merge_speaker(fake_transcript_with_topics):
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -164,7 +157,7 @@ async def test_transcript_merge_speaker(fake_transcript_with_topics):
     assert topics[1]["words"][1]["speaker"] == 0
 
     # merge speakers
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/merge",
         json={
             "speaker_from": 1,
@@ -174,7 +167,7 @@ async def test_transcript_merge_speaker(fake_transcript_with_topics):
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -187,20 +180,19 @@ async def test_transcript_merge_speaker(fake_transcript_with_topics):
 
 
 @pytest.mark.asyncio
-async def test_transcript_reassign_with_participant(fake_transcript_with_topics):
-    from reflector.app import app
-
+async def test_transcript_reassign_with_participant(
+    fake_transcript_with_topics, client
+):
     transcript_id = fake_transcript_with_topics.id
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
     # check the transcript exists
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
     transcript = response.json()
     assert len(transcript["participants"]) == 0
 
     # create 2 participants
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={
             "name": "Participant 1",
@@ -209,7 +201,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert response.status_code == 200
     participant1_id = response.json()["id"]
 
-        response = await ac.post(
+    response = await client.post(
         f"/transcripts/{transcript_id}/participants",
         json={
             "name": "Participant 2",
@@ -219,7 +211,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     participant2_id = response.json()["id"]
 
     # check participants speakers
-        response = await ac.get(f"/transcripts/{transcript_id}/participants")
+    response = await client.get(f"/transcripts/{transcript_id}/participants")
     assert response.status_code == 200
     participants = response.json()
     assert len(participants) == 2
@@ -229,7 +221,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert participants[1]["speaker"] is None
 
     # check initial topics of the transcript
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -246,7 +238,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert topics[1]["segments"][0]["speaker"] == 0
 
     # reassign speaker from a participant
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "participant": participant1_id,
@@ -258,7 +250,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
 
     # check participants if speaker has been assigned
     # first participant should have 1, because it's not used yet.
-        response = await ac.get(f"/transcripts/{transcript_id}/participants")
+    response = await client.get(f"/transcripts/{transcript_id}/participants")
     assert response.status_code == 200
     participants = response.json()
     assert len(participants) == 2
@@ -268,7 +260,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert participants[1]["speaker"] is None
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -285,7 +277,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert topics[1]["segments"][0]["speaker"] == 0
 
     # reassign participant, middle of 2 topics
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "participant": participant2_id,
@@ -297,7 +289,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
 
     # check participants if speaker has been assigned
     # first participant should have 1, because it's not used yet.
-        response = await ac.get(f"/transcripts/{transcript_id}/participants")
+    response = await client.get(f"/transcripts/{transcript_id}/participants")
     assert response.status_code == 200
     participants = response.json()
     assert len(participants) == 2
@@ -307,7 +299,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert participants[1]["speaker"] == 2
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -326,7 +318,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert topics[1]["segments"][1]["speaker"] == 0
 
     # reassign speaker, everything
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "participant": participant1_id,
@@ -337,7 +329,7 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
     assert response.status_code == 200
 
     # check topics again
-        response = await ac.get(f"/transcripts/{transcript_id}/topics/with-words")
+    response = await client.get(f"/transcripts/{transcript_id}/topics/with-words")
     assert response.status_code == 200
     topics = response.json()
     assert len(topics) == 2
@@ -355,20 +347,17 @@ async def test_transcript_reassign_with_participant(fake_transcript_with_topics)
 
 
 @pytest.mark.asyncio
-async def test_transcript_reassign_edge_cases(fake_transcript_with_topics):
-    from reflector.app import app
-
+async def test_transcript_reassign_edge_cases(fake_transcript_with_topics, client):
     transcript_id = fake_transcript_with_topics.id
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
     # check the transcript exists
-        response = await ac.get(f"/transcripts/{transcript_id}")
+    response = await client.get(f"/transcripts/{transcript_id}")
     assert response.status_code == 200
     transcript = response.json()
     assert len(transcript["participants"]) == 0
 
     # try reassign without any participant_id or speaker
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "timestamp_from": 0,
@@ -378,7 +367,7 @@ async def test_transcript_reassign_edge_cases(fake_transcript_with_topics):
     assert response.status_code == 400
 
     # try reassing with both participant_id and speaker
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "participant": "123",
@@ -390,7 +379,7 @@ async def test_transcript_reassign_edge_cases(fake_transcript_with_topics):
     assert response.status_code == 400
 
     # try reassing with non-existing participant_id
-        response = await ac.patch(
+    response = await client.patch(
         f"/transcripts/{transcript_id}/speaker/assign",
         json={
             "participant": "123",

server/tests/test_transcripts_topics.py

@@ -1,22 +1,18 @@
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.asyncio
-async def test_transcript_topics(fake_transcript_with_topics):
-    from reflector.app import app
-
+async def test_transcript_topics(fake_transcript_with_topics, client):
     transcript_id = fake_transcript_with_topics.id
 
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
     # check the transcript exists
-        response = await ac.get(f"/transcripts/{transcript_id}/topics")
+    response = await client.get(f"/transcripts/{transcript_id}/topics")
     assert response.status_code == 200
     assert len(response.json()) == 2
     topic_id = response.json()[0]["id"]
 
     # get words per speakers
-        response = await ac.get(
+    response = await client.get(
         f"/transcripts/{transcript_id}/topics/{topic_id}/words-per-speaker"
     )
     assert response.status_code == 200

server/tests/test_transcripts_translation.py

@@ -1,20 +1,16 @@
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.asyncio
-async def test_transcript_create_default_translation():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post("/transcripts", json={"name": "test en"})
+async def test_transcript_create_default_translation(client):
+    response = await client.post("/transcripts", json={"name": "test en"})
     assert response.status_code == 200
     assert response.json()["name"] == "test en"
     assert response.json()["source_language"] == "en"
     assert response.json()["target_language"] == "en"
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test en"
     assert response.json()["source_language"] == "en"
@@ -22,11 +18,8 @@ async def test_transcript_create_default_translation():
 
 
 @pytest.mark.asyncio
-async def test_transcript_create_en_fr_translation():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post(
+async def test_transcript_create_en_fr_translation(client):
+    response = await client.post(
         "/transcripts", json={"name": "test en/fr", "target_language": "fr"}
     )
     assert response.status_code == 200
@@ -35,7 +28,7 @@ async def test_transcript_create_en_fr_translation():
     assert response.json()["target_language"] == "fr"
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test en/fr"
     assert response.json()["source_language"] == "en"
@@ -43,11 +36,8 @@ async def test_transcript_create_en_fr_translation():
 
 
 @pytest.mark.asyncio
-async def test_transcript_create_fr_en_translation():
-    from reflector.app import app
-
-    async with AsyncClient(app=app, base_url="http://test/v1") as ac:
-        response = await ac.post(
+async def test_transcript_create_fr_en_translation(client):
+    response = await client.post(
         "/transcripts", json={"name": "test fr/en", "source_language": "fr"}
     )
     assert response.status_code == 200
@@ -56,7 +46,7 @@ async def test_transcript_create_fr_en_translation():
     assert response.json()["target_language"] == "en"
     tid = response.json()["id"]
 
-        response = await ac.get(f"/transcripts/{tid}")
+    response = await client.get(f"/transcripts/{tid}")
     assert response.status_code == 200
     assert response.json()["name"] == "test fr/en"
     assert response.json()["source_language"] == "fr"

server/tests/test_transcripts_upload.py

@@ -1,7 +1,7 @@
 import asyncio
+import time
 
 import pytest
-from httpx import AsyncClient
 
 
 @pytest.mark.usefixtures("setup_database")
@@ -14,19 +14,16 @@ async def test_transcript_upload_file(
     dummy_processors,
     dummy_diarization,
     dummy_storage,
+    client,
 ):
-    from reflector.app import app
-
-    ac = AsyncClient(app=app, base_url="http://test/v1")
-
     # create a transcript
-    response = await ac.post("/transcripts", json={"name": "test"})
+    response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["status"] == "idle"
     tid = response.json()["id"]
 
     # upload mp3
-    response = await ac.post(
+    response = await client.post(
         f"/transcripts/{tid}/record/upload?chunk_number=0&total_chunks=1",
         files={
             "chunk": (
@@ -39,14 +36,18 @@ async def test_transcript_upload_file(
     assert response.status_code == 200
     assert response.json()["status"] == "ok"
 
-    # wait the processing to finish
-    while True:
+    # wait the processing to finish (max 10 minutes)
+    timeout_seconds = 600  # 10 minutes
+    start_time = time.monotonic()
+    while (time.monotonic() - start_time) < timeout_seconds:
         # fetch the transcript and check if it is ended
-        resp = await ac.get(f"/transcripts/{tid}")
+        resp = await client.get(f"/transcripts/{tid}")
         assert resp.status_code == 200
         if resp.json()["status"] in ("ended", "error"):
             break
         await asyncio.sleep(1)
+    else:
+        pytest.fail(f"Processing timed out after {timeout_seconds} seconds")
 
     # check the transcript is ended
     transcript = resp.json()
@@ -55,7 +56,7 @@ async def test_transcript_upload_file(
     assert transcript["title"] == "Llm Title"
 
     # check topics and transcript
-    response = await ac.get(f"/transcripts/{tid}/topics")
+    response = await client.get(f"/transcripts/{tid}/topics")
     assert response.status_code == 200
     assert len(response.json()) == 1
     assert "want to share" in response.json()[0]["transcript"]

server/tests/test_webvtt.py

@@ -0,0 +1,151 @@
+"""Tests for WebVTT utilities."""
+
+import pytest
+
+from reflector.processors.types import Transcript, Word, words_to_segments
+from reflector.utils.webvtt import topics_to_webvtt, words_to_webvtt
+
+
+class TestWordsToWebVTT:
+    """Test words_to_webvtt function with TDD approach."""
+
+    def test_empty_words_returns_empty_webvtt(self):
+        """Should return empty WebVTT structure for empty words list."""
+
+        result = words_to_webvtt([])
+
+        assert "WEBVTT" in result
+        assert result.strip() == "WEBVTT"
+
+    def test_single_word_creates_single_caption(self):
+        """Should create one caption for a single word."""
+
+        words = [Word(text="Hello", start=0.0, end=1.0, speaker=0)]
+        result = words_to_webvtt(words)
+
+        assert "WEBVTT" in result
+        assert "00:00:00.000 --> 00:00:01.000" in result
+        assert "Hello" in result
+        assert "<v Speaker0>" in result
+
+    def test_multiple_words_same_speaker_groups_properly(self):
+        """Should group consecutive words from same speaker."""
+
+        words = [
+            Word(text="Hello", start=0.0, end=0.5, speaker=0),
+            Word(text=" world", start=0.5, end=1.0, speaker=0),
+        ]
+        result = words_to_webvtt(words)
+
+        assert "WEBVTT" in result
+        assert "Hello world" in result
+        assert "<v Speaker0>" in result
+
+    def test_speaker_change_creates_new_caption(self):
+        """Should create new caption when speaker changes."""
+
+        words = [
+            Word(text="Hello", start=0.0, end=0.5, speaker=0),
+            Word(text="Hi", start=0.6, end=1.0, speaker=1),
+        ]
+        result = words_to_webvtt(words)
+
+        lines = result.split("\n")
+        assert "<v Speaker0>" in result
+        assert "<v Speaker1>" in result
+        assert "Hello" in result
+        assert "Hi" in result
+
+    def test_punctuation_creates_segment_boundary(self):
+        """Should respect punctuation boundaries from segmentation logic."""
+
+        words = [
+            Word(text="Hello.", start=0.0, end=0.5, speaker=0),
+            Word(text=" How", start=0.6, end=1.0, speaker=0),
+            Word(text=" are", start=1.0, end=1.3, speaker=0),
+            Word(text=" you?", start=1.3, end=1.8, speaker=0),
+        ]
+        result = words_to_webvtt(words)
+
+        assert "WEBVTT" in result
+        assert "<v Speaker0>" in result
+
+
+class TestTopicsToWebVTT:
+    """Test topics_to_webvtt function."""
+
+    def test_empty_topics_returns_empty_webvtt(self):
+        """Should handle empty topics list."""
+
+        result = topics_to_webvtt([])
+        assert "WEBVTT" in result
+        assert result.strip() == "WEBVTT"
+
+    def test_extracts_words_from_topics(self):
+        """Should extract all words from topics in sequence."""
+
+        class MockTopic:
+            def __init__(self, words):
+                self.words = words
+
+        topics = [
+            MockTopic(
+                [
+                    Word(text="First", start=0.0, end=0.5, speaker=1),
+                    Word(text="Second", start=1.0, end=1.5, speaker=0),
+                ]
+            )
+        ]
+
+        result = topics_to_webvtt(topics)
+
+        assert "WEBVTT" in result
+        first_pos = result.find("First")
+        second_pos = result.find("Second")
+        assert first_pos < second_pos
+
+    def test_non_sequential_topics_raises_assertion(self):
+        """Should raise assertion error when words are not in chronological sequence."""
+
+        class MockTopic:
+            def __init__(self, words):
+                self.words = words
+
+        topics = [
+            MockTopic(
+                [
+                    Word(text="Second", start=1.0, end=1.5, speaker=0),
+                    Word(text="First", start=0.0, end=0.5, speaker=1),
+                ]
+            )
+        ]
+
+        with pytest.raises(AssertionError) as exc_info:
+            topics_to_webvtt(topics)
+
+        assert "Words are not in sequence" in str(exc_info.value)
+        assert "Second and First" in str(exc_info.value)
+
+
+class TestTranscriptWordsToSegments:
+    """Test static words_to_segments method (TDD for making it static)."""
+
+    def test_static_method_exists(self):
+        """Should have static words_to_segments method."""
+        words = [Word(text="Hello", start=0.0, end=1.0, speaker=0)]
+        segments = words_to_segments(words)
+
+        assert isinstance(segments, list)
+        assert len(segments) == 1
+        assert segments[0].text == "Hello"
+        assert segments[0].speaker == 0
+
+    def test_backward_compatibility(self):
+        """Should maintain backward compatibility with instance method."""
+        words = [Word(text="Hello", start=0.0, end=1.0, speaker=0)]
+        transcript = Transcript(words=words)
+
+        segments = transcript.as_segments()
+        assert isinstance(segments, list)
+        assert len(segments) == 1
+        assert segments[0].text == "Hello"