This a wyoming protocol server that implements Kokoro using the Kokoro onnx python library.
Recent significant improvements from relvacode:
- Updated to latest Kokoro model
- Available voices are read directly from the model instead of hardcoding
- Support for
SIGINTandSIGTERMtermination - Support for GPU accelerated inference (see ./docker-compose.intel.yml)
- Kokoro debug logging
Docker image avaiable at: https://hub.docker.com/r/nordwestt/kokoro-wyoming
Latest improvements available at:
docker pull ghcr.io/relvacode/kokoro-wyoming:latest
Use one of the provided docker compose files in this repository (use Intel if you have a recent Intel iGPU).
Then go to Home Assistant -> Device & Services -> Add Integration -> Wyoming Protocol. Then, punch in the IP address of where the container is running, and the port (10210) by default.
Once setup, you can use Kokoro as an assistant TTS provider by going to Home Assistant -> Voice assistants and either edit or create an assistant, then select Kokoro as the Text-to-speech provider.
Options for available voices can be found in the model voices documentation.