AudioCraft (MusicGen)
Meta's deep-learning audio generation library bundling MusicGen, AudioGen, and the EnCodec neural codec.
AudioCraft (MusicGen)
AudioCraft is Meta's PyTorch library for deep-learning audio generation and processing. It packages several landmark models together: MusicGen for controllable music generation, AudioGen for text-to-sound, the EnCodec high-fidelity neural codec/tokenizer, and MultiBand Diffusion.
MusicGen is one of the most recognized open music-generation models, producing music conditioned on a text description and an optional melody.
Key features
- MusicGen: text- and melody-conditioned controllable music generation
- AudioGen: text-to-audio environmental and sound-effect generation
- EnCodec: high-fidelity neural audio compression and tokenization
- Training and inference code with pretrained checkpoints
- Simple Python API plus Gradio demos
License note: the AudioCraft code in this repository is MIT; the released model weights are provided under CC-BY-NC (non-commercial), so review each model card before commercial use.
Curated mirror of the open-source AudioCraft (MIT). Get it from the source.