facebookresearch/audiocraft

AudioCraft is a research library from Meta (Facebook Research) that lets you generate audio and music using AI. Give it a text description like "upbeat jazz with piano and drums" and it produces a matching audio clip, no musical knowledge or instruments needed. The library bundles several AI models. MusicGen generates music from text descriptions and can also follow a melody you hum or upload. AudioGen does the same for environmental sounds, things like rain, crowd noise, or footsteps. EnCodec is a neural audio compressor that converts audio into a compact form and back, which the other models use internally. There is also AudioSeal for adding invisible watermarks to AI-generated audio, and JASCO for music generation guided by specific chords, melodies, or drum patterns. Under the hood everything is built on PyTorch, a popular framework for deep learning research. The models are pre-trained, so you can run them without training anything yourself, just install the library and call the model with your text prompt. Training code is also included for researchers who want to fine-tune or build on top of these models. You would use AudioCraft when prototyping apps that need background music generation, when doing audio research, or when experimenting with AI-generated sound design. It requires Python 3.9 and PyTorch. Model weights are available for non-commercial use under a separate license.