AudioCraft

AudioCraft is a codebase for researchers and developers working with generated music, environmental sound, and audio representation. It brings together MusicGen for music, AudioGen for sound generation, and EnCodec for neural audio compression. The product is a library with inference and training resources, so its value comes from technical control and experimentation rather than a turnkey commercial music editor.

Features and what they are useful for

• Generate music conditioned on text, with supported MusicGen workflows also accepting melodic guidance.
• Use AudioGen to explore environmental sounds described in a prompt.
• Work with EnCodec's compressed audio representation as part of an audio research pipeline.
• Inspect inference, training, model cards, and examples within a shared PyTorch-based codebase.

Who it suits

AudioCraft suits audio machine learning researchers, developers prototyping generative audio applications, and creative technologists comfortable with Python. It is especially relevant when you want to inspect or adapt the technical workflow. A musician who simply wants a downloadable track may find a hosted creator product easier to operate.

A practical workflow

For an experimental game prototype, use AudioGen to generate a short outdoor ambience from a description of wind, distant traffic, and footsteps. Listen for unwanted events before importing the result into the game. Separately test MusicGen for a musical direction described by instrumentation and mood. Keep those experiments organized by model, prompt, and configuration so changes can be compared without confusing the music and sound-effect tasks.

Limitations and things to check

Installation and model execution require compatible Python, PyTorch, and supporting dependencies. Hardware requirements differ by model and workload, particularly for training. The repository distinguishes its code license from model-weight licensing, and some weights have noncommercial restrictions. Read the relevant model card and license before treating an experiment as a production asset or embedding the model in a paid service.

Pricing and access

The source code and published model resources can be obtained through the official repository. Local computation, storage, and any hosted execution costs are your responsibility. AudioCraft is not presented as a subscription studio with a bundled commercial soundtrack license; assess the actual code, weights, and use conditions for the model you select.

Frequently asked questions

Are MusicGen and AudioGen the same tool?

They are different models within AudioCraft: one focuses on music and the other on environmental sound.

Can I assume the weights allow commercial use?

No. Review the separate model-weight license and model card before deciding how to use an output or deployment.

Official sources

AudioCraft official website · Official product guide