Echo is a lightweight, asynchronous API service designed for fast and efficient audio transcription. Built on top of Starlette and Faster Whisper, it provides a simple interface to transcribe audio files into text with high accuracy.

• Fast Transcription: Utilizes CTranslate2-powered Whisper models for up to 4x faster inference than original OpenAI Whisper.
• Async API: Built with Starlette for high-performance, asynchronous request handling.
• Configurable: Easily switch model sizes (tiny, base, small, medium, large-v2, etc.) and compute types via environment variables.
• Docker Ready: Includes a Dockerfile for easy containerization and deployment.
• Detailed Output: Returns full transcripts along with segment-level timestamps and language detection probabilities.

• Python 3.11+
• uv (recommended) or pip

1. Clone the repository:

   git clone https://github.com/benpoppy/echo.git
   cd echo

2. Install dependencies:

   uv sync
   # OR using pip
   pip install .

Start the server using Uvicorn:

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

The server will start at http://localhost:8000.

Build and run the container:

docker build -t echo .
docker run -p 8000:8000 echo

You can configure the service using the following environment variables:

Request: GET /health

Response:

ok

Request: POST /transcribe

• Content-Type: multipart/form-data
• Body:
  • file: The audio file to transcribe.

Example using cURL:

curl -X POST -F "file=@/path/to/audio.mp3" http://localhost:8000/transcribe

Response:

{
  "language": "en",
  "language_probability": 0.98,
  "transcript": "Hello world, this is a test transcription.",
  "segments": [
    {
      "start": 0.0,
      "end": 2.5,
      "text": "Hello world, this is a test transcription."
    }
  ]
}

This project relies heavily on these amazing open-source projects:

• Faster Whisper - Faster Whisper transcription with CTranslate2.
• Starlette - The little ASGI framework that shines.