Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning.
MIT License
21.06k
stars
2.17k
forks
source link
CE/PPL values of the default MusicGen small/medium/large models #318
Hi!
I've been retraining MusicGen with a custom dataset of 200,000 audio files (15,000 hours of music) and I wanted to know whats the CE/PPL values of the model trained by the facebook research team.
I'm training a 420.37M param model which is focused on hip-hop music only and the output seems to be good so far, but it still has some ways to be comparable to the default model.
Mainly looking for PPL & CE values of the default small/medium/large model training.
Hi! I've been retraining MusicGen with a custom dataset of 200,000 audio files (15,000 hours of music) and I wanted to know whats the CE/PPL values of the model trained by the facebook research team.
Here are my logs -
I'm training a 420.37M param model which is focused on hip-hop music only and the output seems to be good so far, but it still has some ways to be comparable to the default model.
Mainly looking for PPL & CE values of the default small/medium/large model training.