Piper TTS Docker

A containerized API for text-to-speech conversion using Piper TTS with an OpenAI-compatible endpoint.

Features

• OpenAI TTS API compatible endpoint
• Automatic language detection
• Support for multiple voices
• MP3 and WAV audio formats
• Adjustable speech speed
• Containerized for easy deployment
• API key authentication

Voices

Voice samples are available at: https://rhasspy.github.io/piper-samples/

Download Voices

python3 -m piper.download_voices --download-dir voices da_DK-talesyntese-medium en_GB-alan-medium en_US-amy-medium

Configuration

The following environment variables can be configured:

• API_KEY: Authentication key for the API (default: "CHANGE_ME")
• CUDA_ENABLED: Enable GPU acceleration (default: "false")
• LANGUAGE_VOICE_MAPPING: JSON mapping of language codes to voice models (optional)

Default language mapping:

{
  "da": "./voices/da_DK-talesyntese-medium.onnx",
  "en": "./voices/en_US-amy-medium.onnx",
  "gb": "./voices/en_GB-alan-medium.onnx"
}

Usage

Docker Compose

Use the provided docker-compose files to start the service:

# Development
docker-compose up
# Server deployment
docker-compose -f docker-compose.server.yml up -d

API Endpoints

Health Check

GET /health

Generate Speech

POST /audio/speech

Request body:

{
  "model": "tts-1",
  "voice": "en",
  "input": "The text you want to convert to speech",
  "response_format": "mp3",
  "speed": 1.0,
  "auto_detect_language": true
}

Authentication: Bearer token with your configured API_KEY

Note: That the model and voice do not have any effect at the current release. They are there to be compatible with openAI's api.