PABannier / bark.cpp

Suno AI's Bark model in C/C++ for fast text-to-speech
MIT License
630 stars 48 forks source link

Support for piper models #135

Open danemadsen opened 3 months ago

danemadsen commented 3 months ago

It would be helpful to add support for piper models into bark.cpp

there is already a c++ library for piper but it is difficult to compile and does not work well cross platform. Piper is currently running on the onnx runtime.

https://github.com/rhasspy/piper

PABannier commented 2 months ago

Hi @danemadsen ! The piper model looks indeed really nice. Would you like to try to implement it? I'm happy to guide you in the process.

danemadsen commented 2 months ago

Hi @danemadsen ! The piper model looks indeed really nice. Would you like to try to implement it? I'm happy to guide you in the process.

Haha, I'd love to help implement it but I'm afraid I'm not educated on the topic to be able to pull it off.

I've looked into vits.cpp (an incomplete GGML implementation of vits / piper models) , it compiles but crashes instantly.

My understanding is the main challenge of a vits implementation will be finding a way to handle phonemization without using espeak-ng.

I'm wanting an implementation without espeak-ng as I'm wanting to use a vits model in my llama.cpp android frontend maid and espeak-ng struggles to compile for android.

Thanks for any advice.