HAxTTS
This is not an integration, but rather a separate server with compatibility for Mary-TTS.
It is pretty simple solution with inference optimizations (thx to the Auralis project) for running XTTS models on GPU. My solution is currently supported only by Mary-TTS integration. Maybe in the future I will add support for Wyoming Protocol. The main advantage over AllTalk TTS and others is multithreaded synthesis, which allows to significantly reduce the time of speech synthesis, especially on large volumes of text.
For quick start:
- Docker
docker run -p 9898:9898 --gpus all --name hax-tts-service -d alekst7/haxtts-service:latest - In Home Assistant in the configuration.yaml:
tts:
- platform: marytts
host: localhost # IP address of the docker server
port: 9898
codec: WAVE_FILE
voice: xenia # The name of the voice you want to use.
language: en_US # The model is multilingual, it only affects the pronunciation accent.
I’ll be glad to receive feedback and wishes.