Fine tuning support

[GUUser91]

This fork of bark includes fine tuning support. Can you add this feature in?
https://github.com/serp-ai/bark-with-voice-clone

[matbeedotcom]

training doesnt work because suno wont release the weights

[JonathanFly]

I have tested that fine-tuning just briefly. It's challenging to get it just right. I'm for sure going to have something along those those lines. You're going to be able to use large samples of voice sample and do fun things with them. I've had more luck combining very long inferencing with very light fine-tuning, then the fine-tuning by itself.

I know I haven't updated Bark Infinity forever, but it's not dead. I've have been adding to my local fork as I try out experiments. The problem is the endless experimentation made using the code for 'real work' difficult or even broken and cleaning it up is going to take a while. I have to just bypass most of my own code to test against default Bark quite often because my sampling is so convoluted. I also want to move over the Hugging Face Bark implementation but that's turning out to be more challenging than I expected with all my custom changes, and understanding how to port everything over to their conventions and API is going to take me awhile to wrap my head around.

I don't know if anyone else cares, but finally, full colored terminal output. Useful? No. But it just makes my day better.

If I had to guess it'd be still 3 weeks till I update, pretty busy with regular work too.

[platform-kit]

@JonathanFly did you ever make any progress with the fine-tuning script?