Google allows users to create music using the MusicLM neural network

Google has announced that the MusicLM neural network, which composes music based on text descriptions, is now available to the general public. Information about this came from TechCrunch with reference to the Google I/O 2023 presentation.

The MusicLM neural network was introduced in January 2023 and is trained on 280,000 hours of recordings, making it capable of generating songs in any genre. At the time, Google announced that it would not make the technology widely available, but now the company has decided to change its mind and put MusicLM in the public domain.

The MusicLM neural network is included in the AI Test Kitchen package for iOS and Android. To get access to this technology, you need to request to join the testing program on a special website.

MusicLM successfully combines genres and instruments, and writes music based on abstract queries. For example, in January testing, the neural network successfully combined dance music and reggaeton, created a composition for the query "feeling of awe and reverence" and composed a track with a "cosmic and otherworldly" sound.

In addition, the artificial intelligence creates two variants of a piece of music based on a textual user query so that the user can choose the more appropriate variant. The text description can contain information not only about the instrumental features of the composition, but also about the emotional mood of the music. The neural network understands only English for now, but it is possible that the company will expand the language set in the future.

Google officials said. MusicLM will not generate compositions using the vocals of specific artists, due to possible copyright issues. However, the neural network is capable of generating music with vocals.