Description

Model was trained from 17.5 minutes of _clean_ 768Kbps WAV files, meaning the datasets are not from an episode or a song with post-processing. This ensures that the model could replicate as close to the actual raw recording as possible. "A model trained from good datasets beats a model trained from subpar datasets, no matter how much you trained it" Anyways here's the demo all were made on mangio-crepe with 120 hop _long live make-heroines_