Meta Launches AudioCraft, An AI Tool That Can Create Music Based On Text

JAKARTA - Meta has just released a generating set of Artificial Intelligence (AI) tools to make music and audio from text requests, called AudioCraft.

"Imagine a professional musician being able to explore a new composition without having to play a single tone on the instrument," Meta said in its official blog, quoted Thursday, August 3.

"Or an indie game developer who fills cyberspace with realistic voice effects and surrounding noise with a limited budget. That's AudioCraft's promise."

AudioCraft consists of three models, including AudioGen which is a tool to produce various audio and soundscapes effects, MusicGen which can create musical compositions and melodies from descriptions or texts, and finally EnCodec is a neural network-based audio compression codec.

MusicGen itself was trained on about 400,000 recordings along with text descriptions and metadata, amounting to 20,000 hours of music owned by Meta or licensed specifically for this purpose.

According to Meta, currently the generative AI model is mostly centered around text and stationary images, but the development of a generative audio tool has lagged behind.

"There's some work out there, but it's very complicated and not too open, so people can't easily play it," Meta said.

Therefore, Meta creates AudioCraft which is indeed used for the manufacture and compression of music and sound in one roof.

The company hopes that the AudioCraft release under the MIT License will contribute to a wider community by providing accessible tools for audio and music experiments.

"By sharing the code for AudioCraft, we hope other researchers can more easily test a new approach to limiting or eliminating potential bias and abuse of a generating model," explains Meta.