Kyutai TTS Server

Kyutai TTS Server is an OpenAI-compatible Text-to-Speech (TTS) API server built with FastAPI. It provides a simple interface for generating speech from text using state-of-the-art TTS models.

Features

• OpenAI-compatible API for text-to-speech generation
• Supports multiple audio formats (WAV, MP3, FLAC, OGG)
• Multi-voice and Multi Dialogue generation support
• Health check endpoint
• Docker support for easy deployment

Voice Library

The Kyutai (voices)[https://huggingface.co/kyutai/tts-voices] will be automatically downloaded, along with the (model)[https://huggingface.co/kyutai/tts-1.6b-en_fr] itself, from huggingface during startup to the cache directory.

For the API, you can provide the wav file path from this voice library. (for example expresso/ex03-ex01_happy_001_channel1_334s.wav).

Project Structure

Kyutai-TTS-Server/
├── app/
│   ├── __init__.py
│   ├── server.py
│   ├── config.py
│   ├── models.py
│   ├── tts.py
│   └── utils.py
├── .gitignore
├── docker-compose.yaml
├── Dockerfile
├── requirements.txt
└── README.md

API Endpoints

Text-to-Speech Generation

• URL: /v1/audio/speech
• Method: POST
• Request Body:

  {
    "model": "optional-model-name",
    "input": "Text to convert to speech",
    "voice": "expresso/ex03-ex01_happy_001_channel1_334s.wav",
    "response_format": "wav|mp3|flac|ogg",
    "speed": 1.0
  }

• Response: Audio file in the requested format

Multi-Voice, Multi-Dialogue Text-to-Speech Generation (extended from OpenAI API)

• URL: /v1/audio/speech
• Method: POST
• Request Body:

  {
    "model": "optional-model-name",
    "inputs": [
      "Hey there, I'm speaker A",
      "And I'm speaker B!",
      "We are having a dialogue"
    ],
    "voices": [
      "expresso/ex03-ex01_happy_001_channel1_334s.wav",
      "expresso/ex04-ex02_sarcastic_001_channel2_466s.wav"
    ]
    "response_format": "wav|mp3|flac|ogg",
    "speed": 1.0
  }

• Response: Audio file in the requested format

Health Check

• URL: /health
• Method: GET
• Response:

  {
    "status": "healthy"
  }

Running the Server

Using Docker

1. Build the Docker image:

   docker build -t kyutai-tts-server .

2. Run the Docker container:

   docker run -d -p 8000:8000 kyutai-tts-server

Locally

1. Install the required dependencies:

   pip install -r requirements.txt

2. Run the server:

   python server.py --host 0.0.0.0 --port 8000 --reload

License

This project is licensed under the MIT License.