Meta releases sound generation tool AudioCraft

Meta has released an open source AI tool called AudioCraft, allowing users to generate “realistic” music and environmental sounds based on text prompts.

AudioCraft works for music and sound generation, as well as compression, according to Meta. It consists of three models, namely, MusicGen, AudioGen, and EnCodec.

MusicGen was released in June and was trained on 20,000 hours of licensed music. It generates samples based on both textual and melody prompts.

AudioGen was trained on public sound effects and generates audio from text-based inputs. Its updated AudioCraft version “lets you generate environmental sounds and sound effects like a dog barking, cars honking, or footsteps on a wooden floor,” Meta said.

EnCodec is a decoder designed to improve the quality of generated music and was also updated for the AudioCraft release.

“The models are available for research purposes and to further people’s understanding of the technology,” Meta said in a blog post, adding that researchers and practitioners will be able to train their own models based on their own datasets for the first time.

It said that the tool would allow independent game developers to create realistic sound effects “on a shoestring budget” and professional musicians to “explore new compositions without having to play a single note on an instrument.”

Earlier this year, Google released its own text-to-music generator called MusicLM, as creators continue to express concern over the proliferation of AI-generated music and its impact on the industry.