Neural network MusicLM from Google to compose music from a text description

MusicLM, like all popular models, is not a full-fledged AI – these are vulnerable neural networks that do not create new content, but select fragments from existing audio recordings to meet the requirements. She has a very good source data discovery system, thanks to which she understands the specific requirements of users and skillfully selects a wide variety of musical styles. The problem is that Google has disabled MusicLM’s maximum power to reduce the risk of copyright infringement.

Google’s decision is more than justified, because experienced musicians agree by ear, where they get into certain fragments, rhythms, guitar riffs, etc. The neural network skillfully selects musical inserts that together sound like Odie’s generated melody, but at the same time do not make any sense – like drumming in acoustics with a violin. It perfectly imitates the sound of musical instruments, musical style and trends in music, unnoticed full-fledged compositions gives out a skillful and almost always meaningless cacophony of sound, although people in the morning are unlikely to catch it.

At the same time, professional musicians have noted that MusicLM easily allows important instructions if it cannot implement them. The possibilities of a neural network are not endless, and built-in limitations force it to generate quick results, rather than work for hours to find effective solutions. Therefore, MusicLM can hardly compete with human musicians, but can become a source of inspiration, thanks to its ability to create musical improvisations for any was afraid.