implement example_rvc with TextToAudioStream rather than BufferStream

[Meshwa428]

can you please update the example_rvc/xtts_rvc_synthesizer.py to utilize TextToAudioStream rather than using BufferStream.

because for cpu users, the audio is very oddly being played.

Recording.2024-10-06.172841.mp4

as you can see in the above video

[KoljaB]

This is very probably because CPU inference is too slow to do realtime XTTS synthesis. Even a lot of GPUs are. I don't think BufferStream has anything to do with is.

[Meshwa428]

But I can do real time inferencing using the coqui engine and there is no interference like this one is having. When I use the Coqui engine, it runs completely fine.

I am not using XTTS either, the one I am using is way smaller than that

It's tts_models/en/vctk/vits

Isn't there a way to fix this?

Maybe the issue is that we are applying rvc on the audio chunk rather than full audio of the whole sentence