Google introduced a neural network that generates music from a text description

Google announced MusicLM is the latest generative artificial intelligence that can turn textual descriptions of varying complexity into quality music.

The developers note that text-to-music conversion models are not new, but MusicLM is superior to previous systems in both sound quality and adherence to textual description.

The neural network was trained on 280 thousand hours of music so that it could cope with complex queries. All melodies are created at a frequency of 24 kHz, which remains unchanged for a couple of minutes.

In the given examples the Google site presents 30-second snippets generated from just a couple of sentences.

MusicLM is capable of generating music from various genres and even playing "musical experience level" (for example, beginner or professional). In the future, Google may work on creating lyrics, improving the quality of vocals and upsampling.

So far, the developers do not plan to upload MusicLM to the public.