A high-performance, production-ready speech-to-text API service based on the faster-whisper library.

• Efficient transcription using the CTranslate2-based faster-whisper implementation
• Support for all Whisper models (tiny, base, small, medium, large-v2, large-v3, distil)
• REST API with comprehensive endpoints
• Background task processing
• Word-level timestamps
• Voice Activity Detection (VAD) for filtering silence
• Batched inference for faster processing
• Docker and Docker Compose support for easy deployment
• CUDA support for GPU acceleration

• Docker and Docker Compose (recommended)
• NVIDIA GPU with CUDA support (optional but recommended for performance)
• Python 3.8+ (if not using Docker)

1. Clone this repository:

   git clone https://github.com/msjd78/Whisper-FastAPI-Transcription-Service.git
   cd Whisper-FastAPI-Transcription-Service

2. Start the service:

   docker-compose up -d

3. The service will be available at http://localhost:8000.

1. Install dependencies:

   pip install -r requirements.txt

   or

   pip install fastapi uvicorn python-multipart faster-whisper pydantic

2. Start the service:

   uvicorn app:app --host 0.0.0.0 --port 8000 --reload

After starting the API, you can access the UI at:

http://localhost:8000/static/index.html

This provides a user-friendly interface for transcription.

POST /api/transcribe

Upload an audio file for transcription. Returns a task ID that can be used to check the status.

• Content-Type: multipart/form-data
• Body:
  • file: Audio file (mp3, wav, etc.)
  • options (optional): JSON string with transcription options

{
  "model_size": "large-v3",
  "device": "cuda",
  "compute_type": "float16",
  "language": "en",
  "batch_size": 16,
  "beam_size": 5,
  "word_timestamps": true,
  "vad_filter": true,
  "vad_parameters": {
    "min_silence_duration_ms": 500
  },
  "condition_on_previous_text": true,
  "use_batched_mode": true
}

GET /api/tasks/{task_id}

Check the status of a transcription task.

GET /api/tasks?limit=10&status=completed

List transcription tasks with optional filtering.

DELETE /api/tasks/{task_id}

Delete a transcription task and its associated data.

GET /api/health

Check if the service is running properly.

A client script is provided for testing the API:

python client.py http://localhost:8000 path/to/audio.mp3 output.txt

The service can be configured using environment variables:

• PORT: Port to run the service on (default: 8000)
• HOST: Host to bind to (default: 0.0.0.0)
• MODEL_DIR: Directory to store downloaded models (default: None)

For best performance:

1. Use GPU acceleration with CUDA
2. Use the batched mode for faster processing
3. Choose the appropriate model size for your needs:
   • tiny, base, small: Lower quality but faster
   • medium: Good balance between quality and speed
   • large-v3: Highest quality but slower
   • distil-large-v3: Comparable to large with faster processing time

MIT License