[feature request] add trained model with saved weights

[amirgamil]

Would be great if the repo also included a trained version of this model with weights people can just download and use :)

[ckwdani]

After the training is stable for Soundstream, I can share my weights for it :)

[turian]

@BlackFox1197 what corpus are you training on?

[ckwdani]

@BlackFox1197 what corpus are you training on?

Im Training on Librispeach.

[turian]

@BlackFox1197 Would be wonderful if you would train on audioset since that is the most diverse corpus. Or FSD50K which is smaller but diverse. One concern I have is that soundstream reported poor results on music, as opposed to encodec

That's soundstream, you can see the music numbers (green) are all worse.

Versus encodec:

[lancer1256]

@BlackFox1197 Are the model weights ready to go? :)