Meta has released its own AI-powered music generator – and, unlike Google, open-sourced it, TechCrunch reported.
Called MusicGen, Meta’s music-generating tool, can turn a text description (e.g. “An ‘80s driving pop song with heavy drums and synth pads in the background”) into about 12 seconds of audio, give or take. MusicGen can optionally be “steered” with reference audio, like an existing song, in which case it will try to follow both the description and melody.
Meta says that MusicGen was trained on 20,000 hours of music, including 10,000 “high-quality” licensed music tracks and 390,000 instrument-only tracks from ShutterStock and Pond5, a large stock media library. The company hasn’t provided the code it used to train the model, but it has made available pre-trained models that anyone with the right hardware – chiefly a GPU with around 16GB of memory – can run.
So how does MusicGen perform? TechCrunch said – certainly not well enough to put human musicians out of a job. It’s songs are reasonably melodic, at least for basic prompts like “ambient chiptunes music.” Writer Kyle Wiggers said the music is: on par (if not slightly better) with the results from Google’s AI music generator, MusicLM. But they won’t win any awards.
According to TechCrunch, it might not be long before there’s guidance on the matter. Several lawsuits making their way through the courts will likely have a bearing on music-generating AI, including one pertaining to the rights of artists whose work is used to train AI systems without their knowledge.
Music:)ally reported that MusicGen is described as “a simple and controllable music generation LM [language model] with descriptions of the music you’d like it to create, and it whips up 12-second samples in response.”
The first question for many rights holders will be: how was this trained. That’s explained in the accompanying academic paper.
“We use 20k hours of licensed music to train MusicGen. Specifically, we rely on an internal dataset of 10k high-quality music tracks, and on the ShutterStock and Pond5 music data” – referring to the popular stock-music libraries.
Meta joins other technology companies in developing (and releasing for public consumption) AI-music models. Alphabet recently unveiled its MusicLM, trained on around 280,000 hours of material from the Free Music Archive, and made it available for people to test out.
According to music:)ally, the music AI-models developed by OpenAI, Alphabet, and now Meta are research projects rather than commercial products at this point.
They’re more likely to become the basis for startups and developers to use than they are to signify a serious move into AI music by the bigger companies.
In my opinion, all of this is fine, until one of these AI music makers creates something that sounds like a Metallica song.