Meta state-of-the-art open-source audio deep learning framework unifying MusicGen, AudioGen, and EnCodec
100% free and open-source PyTorch library for unrestricted local GPU deployment and research
AudioCraft is Meta premier open-source deep learning library for generative audio and sound synthesis. It unites three cutting-edge neural models: MusicGen (generates high-fidelity music conditioned on text prompts or reference melodies), AudioGen (synthesizes realistic Foley sound effects, environmental soundscapes, and ambiances from descriptive text), and EnCodec (state-of-the-art neural audio compression codec). Researchers and developers can inspect, train, and run models locally with complete autonomy.