Wyoming IndexTTS - Voice Cloning TTS for Home Assistant

A Docker service implementing the Wyoming protocol for Home Assistant, powered by IndexTTS-2 zero-shot voice cloning.

Drop any number of voice WAV clips into a directory and they each become a selectable voice in Home Assistant's TTS dropdown.

tts-experiment/
├── Dockerfile
├── docker-compose.yml
├── voices/                      # Drop your .wav files here
│   ├── morgan.wav               # Each file = a selectable voice
│   └── sarah.wav
├── scripts/
│   └── run.sh                   # Container entrypoint
└── wyoming_indextts/
    ├── __init__.py
    ├── __main__.py              # Wyoming server setup
    └── handler.py               # TTS synthesis handler

• Implements the Wyoming protocol (TCP on port 10300) so Home Assistant discovers it as a TTS provider
• Uses IndexTTS-2 for zero-shot voice cloning — give it any reference WAV and it synthesizes speech in that voice
• All .wav files in the voices directory are automatically registered as selectable voices
• The TTS model is preloaded at startup so the first request is fast (no cold start)
• Model checkpoints are auto-downloaded from HuggingFace on first run and persisted in a Docker volume

1. Add your voice files to the voices/ directory (any short WAV clips of voices you want to clone):

   voices/
   ├── morgan.wav
   └── sarah.wav
   

2. Build and run:

   docker compose up -d

3. Add to Home Assistant via Settings > Devices & Services > Add Integration > Wyoming Protocol, entering your host IP and port 10300

4. Select a voice in Settings > Voice Assistants > pick your pipeline > TTS voice dropdown

All configuration is done via environment variables in docker-compose.yml:

Drop new .wav files into voices/ and restart:

docker compose restart

The filename (without .wav) becomes the voice name in Home Assistant. For example, voices/morgan.wav appears as "morgan" in the voice dropdown.

To set a specific default voice:

environment:
  - DEFAULT_VOICE=morgan

• NVIDIA GPU with CUDA 12.8+ support
• NVIDIA Container Toolkit installed