Add RVC support

[lainedfles]

Consider https://github.com/Vali-98/XTTS-RVC-UI for inspiration or possible merge.

[240db]

Consider https://github.com/Vali-98/XTTS-RVC-UI for inspiration or possible merge.

@lainedfles the XTTS-RVC-UI project unfortunately suffers from dependency hell with a lot of users struggling to install it. I could not get it working whatsoever.
Vali-98/XTTS-RVC-UI#8

The repo's last update was one year ago, so it definitely needs to be revised given all those issues.
I have seen some new posts there that suggest a few other things, but I have tried most. I will give it a try again.

The only RVC project I got to work on my Linux system was https://github.com/240db/Retrieval-based-Voice-Conversion-WebUI
which was also broken because they were using a buggy gradio version. My fork actually updated that so the install can be much easier. But that project is not compatible with the xtts models, so you have to train from scratch, I think it uses VITS instead of xtts.

But your idea is great, currently I train my models using this xtts-finetune-webui repo, then I do inference using the manual API, and created a simple gradio app that allows the user to select different models from the dropdown bar, one of the things that is missing is RVC because it would be great to guide the intonation for specific words or phrases that the model struggles to reproduce.

I will give this a try again, if I can get it working I will fork with the necessary changes and we can easily incorporate RVC in the inference tab.

[lainedfles]

Perfect, I'd planned to make my own attempt eventually. My hope is that the project mentioned in #82 might enable this potential. It looks to be under active development so maybe it will become a realistic future replacement for Cocui XTTS. Despite caveats (space duplication, network complexity, etc.), container based deployment of CUDA/Python is one of the best strategies when working with an unforgiving dependency chain.