• by Colin Kirkland , June 12, 2023

On the heels of Google’s AI music generator MusicLM, which was released last month, Meta has launched a demo of “MusicGen,” the social giant's open-source text-to-audio generative AI tool that can turn “an 80s driving pop song with heavy drums and synth pads in the background” into about a dozen seconds of music.

In a series of tweets, Meta AI researcher Felix Kreuk described MusicGen as “a simple and controllable music generation model,” adding that the company is releasing code and models “for open research, reproducibility, and for the music community.”

According to Meta, MusicGen was trained on 20,000 hours of music, including 10,000 “high-quality” licensed tracks and 390,000 instrumental tracks from stock media libraries ShutterStock and Pond5.

To increase the program's speed, Meta used its 32Khz EnCodec audio tokenizer to generate small bits of music that can be processed in parallel. The generator can also be “steered” with existing songs, creating a new clip with both description and reference audio.

MusicGen is open-source, setting it apart from Google’s MusicLM, and can be used to generate commercial music or anything else.

Meta is not imposing restrictions on how the program can be used, stating that the music MusicGen was trained on is “covered by legal agreements with the right holders.”