LostRuins / koboldcpp

Run GGUF models easily with a KoboldAI UI. One File. Zero Install.
https://github.com/lostruins/koboldcpp
GNU Affero General Public License v3.0
5.32k stars 363 forks source link

Oute TTS integration #1202

Closed Raskoll2 closed 2 weeks ago

Raskoll2 commented 2 weeks ago

https://github.com/edwko/OuteTTS/

Oute TTS is a llama based speech model, meaning it works with llama.cpp and can be quantized as a gguf. This would be an awesome model to have in the audio tab like whisper.

I don't think koboldcpp supports the TTS part of the openai API, and implementing it entirely for this model would suck, but just adding it as a drop down TTS provider is plenty. Having a standard voice pipeline would be a good base to create things like OpenWeb UI's call mode, and home assistant integration or what not. It opens doors!

Being llama.cpp compatible should make this pretty easy, please consider it 🙏. Thanks for your hard work

LostRuins commented 2 weeks ago

It's not llama.cpp compatible.

OuteTTS uses and relies on pytorch, not ggml. Take a look at https://github.com/edwko/OuteTTS/blob/main/outetts/v0_1/encoder/model.py