Meta Platforms Inc., formerly known as Facebook, Inc., has recently unveiled a new AI tool for its users. This new generative AI, MusicGen, is available on a third-party website called HuggingFace.
It has been reported that Meta released MusicGen, a generative music tool, just a month after Google launched a similar tool called “musicLM.” MusicGen is considered a strong contender in the growing market for generative AI.
According to Felix Kreuk, an AI developer for Meta AI, MusicGen is “A simple and controllable music generation model.”.
MusicGen by Meta utilizes a transformer language model and has been trained on 20,000 hours of licensed music. Most of the music used as training data comes from sources such as Shutterstock and Pond5.
As a result, this tool can create new music from text or audio samples. Users can describe the music they want to listen to or supply an audio clip. Using its foundation model, MusicGen can create 15 seconds of original audio.
Explaining the decision to make MusicGen open-sourced, Felix stated on Twitter, “We release code (MIT) and models (CC BY-NC) for open research, reproducibility, and for the music community.”
Interested users can log into HuggingFace API and use the tool for creative purposes. Additionally, those who have signed up for Hugging Face Space can generate 120 seconds of music with this tool.