What it does
AudioCraft is a code base from Meta AI for generative audio, covering music, sound effects and audio compression learned from raw audio signals. The site presents it as AI research for audio. MusicGen produces music from text input, and AudioGen produces environmental sounds from text. Both are single autoregressive language models that work over streams of discrete audio tokens. The EnCodec neural audio codec learns those tokens from the raw waveform, and its decoder maps generated tokens back into audio. Different conditioning models, such as a pretrained text encoder, control what is generated. The site offers listening samples for each model and links to the code on GitHub, model cards, technical details and a Meta AI blog post. It does not state pricing or supported platforms.
Capabilities
- MusicGen for generating music from text
- AudioGen for generating environmental sounds from text
- EnCodec neural audio codec that turns audio into discrete tokens
- Single autoregressive language model design shared by MusicGen and AudioGen
- Code and model cards published on GitHub
When to use it
- Generating music clips from a text description
- Generating environmental sound effects from text
- Studying or building on audio generation research code
