Google’s MusicLM neural network will compose music based on text descriptions

Google has opened access to a new service called MusicLM, which is a kind of music generator. It is positioned as a full-fledged digital composer based on artificial intelligence, which can compose any melody based on a test description.

MusicLM, like all popular analogues, is not a full-fledged AI – it is a kind of neural network that does not create new content, but selects fragments from existing audio recordings to match the request. It has a very good system of initial data recognition, so it understands the nuances of users’ requirements and skillfully varies a variety of music styles. The problem is that Google has shut down MusicLM’s core capabilities to avoid copyright infringement charges.

Google’s decision is more than justified, because experienced musicians can tell by ear where certain fragments, rhythms, guitar riffs, etc. are borrowed from. The neural network skillfully selects musical inserts that fit into the general sound of the generated melody, but do not make any sense – like a drumbeat combined with a violin. It perfectly imitates the sound of dozens of instruments, musical styles and trends in music, but instead of full-fledged compositions produces an artful and almost always meaningless cacophony of sounds, although ordinary people are unlikely to understand it.

At the same time, professional musicians noted that MusicLM easily ignores important instructions if it cannot implement them. The possibilities of a neural network are not infinite, and the built-in limitations force it to generate a quick result, rather than laboring for hours to find the best solution. Therefore, MusicLM is unlikely to compete with human musicians, but it can be a source of inspiration, thanks to its ability to create musical improvisations on any theme.

You may be interested: Biden is waiting for Congress to take action to tighten gun control