Google has created a system called MusicLM that can create music in any genre from a text description. Unlike other AI music generators like Google’s AudioML and OpenAI’s Jukebox, MusicLM has a large model and training database (280,000 hours of music) that allow it to produce music with amazing variety and depth.
MusicLM can combine genres and instruments and write tracks using abstract concepts that are normally difficult for computers to understand.
But not only that! You can also create melodies based on singing, whistling, or describing a painting. It also features a story mode that combines various descriptions to produce a DJ set or soundtrack.
MusicLM problems
However, MusicLM also has its problems. Some compositions sound strange and the vocals tend to be incomprehensible. Although the performances are better than one would expect, they can be repetitive in ways that human works are not.
On the other hand, due to copyright concerns, Google researchers they will not release MusicLM to the public. At the time of publication, about 1% of the music produced was a direct copy of the training songs. Although licensing issues for AI music have yet to be resolved, a 2021 report by Eric Sunray suggests that there are enough “coherent traces” of the original sounds that AI music may violate copyright. Releases may need to be obtained to release songs created by AI, just like musicians who rely on samples. But hey, if he you you you Sharika’s passed, surely MusicLM will pass too.
AI already has a place in music. Artists like Holly Herndon and Arca have used algorithms to produce albums and soundtracks. However, MusicLM may not be ready for the public, but it suggests a future where AI could play a bigger role in the studio.
Link: google-research.github.io/seanet/musiclm/examples/
