Amid the race for artificial intelligence (AI)-powered software, Google has unveiled on GitHub the MusicLM, a neural network capable of generating full musical compositions based on user text queries.The company, however, has no plans to make it available to the public in the near future.
As Business Insider reports, a media library with a total duration of 280,000 hours of audio files was used to train the MusicLM AI.
In a research paper published, Google researchers described MusicLM as "a model generating high-fidelity music from text descriptions such as 'a calming violin melody backed by a distorted guitar riff.'"
Also, it is noted that MusicLM can create compositions both based on specific requests, with the performance of specific musical instruments, and based on abstract descriptions of mood or feelings.
In order to demonstrate the capabilities of the neural network, Google has published dozens of tracks. These include Music for an Arcade Game, Music for Lost in Space, Melody for Meditation, Vincent van Gogh's Starry Night, Salvador Dali's The Persistence of Memo, and more.
The company has indicated that it has no plans to make MusicLM publicly available yet. One reason is the potential for copyright infringement. About 1% of the creations generated by the neural network are completely duplicates of licensed music taken from the library for educational purposes.
Google has cited other risks, too, as to why this service will not be available to the public yet; namely programming bias that could lead to inappropriate representation and appropriation of cultural values, technological disruptions, etc.
"We strongly emphasize the need for more future work in tackling these risks associated to music generation — we have no plans to release models at this point," the aforesaid study states.