Alternative Local TTS Engine Support

[slender cloud]

Hey guys! I'm setting up a smart home for my parents, and I need some help with the voice assistant. You see, they don't speak English well, only Bulgarian. Because of that, I decided to go with a Whipser-Turbo instance for STT, Gemini 2.5 Flash for LLM inference with Qwen3-4b as a fallback in case the internet goes out, and the only thing left is the TTS. Unfortunately, I can't find much info on the HA Voice documentation on using alternative local TTS engines than Piper. Sadly, Piper doesn't support Bulgarian. I like how mms-tts-bul performs, and there's probably a few other good tts engines, but I'm not sure if it's possible to use alternatives with Home Assistant Voice. Has anyone else tried it/done it before? do I need anything specific to implement it?

[dark zodiac]

Search for Wyoming versions of TTS engines, they will work with HA. Like "kokoro Wyoming"

[slender cloud]

hm. I can't seem to find mms-tts wyoming. is there a way for me to "manually" turn the input/output into that required by the Wyoming protocol?

[dark zodiac]

Nope..

[topaz shoal]

If you have basic programming experience, you can easily add protocol support to any TTS with gemini or grok. This integration can be used as a starting point - https://github.com/tarocco/wyoming-say/tree/main/wyoming_say. It's pretty straightforward and well commented on. It is also not necessary to use wyoming for voice. Any tts configured in HA can be used for speech synthesis.
If you are satisfied with the cloud service - there is such integration https://github.com/hasscc/hass-edge-tts/

[slender cloud]

oh awesome!

I'm not exactly sure of the api protocols used for TTS, but if you say it can be done, that's a relief 🙂
I'll probably set up a python listener waiting for API requests from Home Assistant, pass that onto the local TTS model, and return the wav file in whatever format HA expects