Which selfhosted TTS provider should I use with home assistant?

Submitted ⁨⁨1⁩ ⁨week⁩ ago⁩ by ⁨scrubbles@poptalk.scrubbles.tech⁩ to ⁨selfhosted@lemmy.world⁩

Title question mostly. I’ve played with XTTS-v2 and it worked pretty well, but I’m wondering if folks are using anything else special. I’d like to train my own voice finetune which is what I did with XTTS-v2, and then use it with home assistant’s voice feature. Welcome all opinions on it!

source

Comments

Sort:hotnew top

ALERT@sh.itjust.works ⁨1⁩ ⁨week⁩ ago
if you need English - right now it’s kokoro-fastapi github.com/remsky/Kokoro-FastAPI set this container up and use it as an openai TTS endpoint using this hacs integration github.com/sfortis/openai_tts

source
- scrubbles@poptalk.scrubbles.tech ⁨1⁩ ⁨week⁩ ago
  Very nice! I’ll check this out!
  
  source
johntash@eviltoast.org ⁨1⁩ ⁨week⁩ ago
Piper works pretty well. I’m only using it because it was easier to find a custom glados voice.

Kokoro has good default voices. I also started trying out Speaches recently. It provides an open ai api wrapper around several options

source
- VocalCritic@lemmy.ml ⁨1⁩ ⁨week⁩ ago
  Any tips on getting speaches to work with Home assistant? Got speaches working but haven’t gone the next step yet.
  
  source
just_another_person@lemmy.world ⁨1⁩ ⁨week⁩ ago
Pretty much just personal preference at this point. XTTS is certainly not the most efficient though.

source
- scrubbles@poptalk.scrubbles.tech ⁨1⁩ ⁨week⁩ ago
  any personal preferences you recommend?
  
  source
  - just_another_person@lemmy.world ⁨1⁩ ⁨week⁩ ago
    Pico, Mary, and Google all run locally and off of CPU only.
    
    I think all the rest require cloud accounts or acceleration hardware to work quickly.
    
    I’m personally fine with Mary or Pico, but I know some people like the fancier ones.
    
    source
    -> View More Comments
HybridSarcasm@lemmy.world ⁨1⁩ ⁨week⁩ ago
You might consider asking this in the !homeassistant@lemmy.world community, too.

source
Vector@lemmy.world ⁨1⁩ ⁨week⁩ ago
Don’t know much about the training side of things, but I have Piper set up with home assistant using the Wyoming protocol and it just goes. Some of the out-of-the-box voices are pretty decent too.

source