Google has created the MusicLM neural network, which generates music of any genre from text descriptions

Google has developed a new neural network called MusicLM. But it is not going to release it yet.

Details

It is an artificial intelligence that generates music from a text description, and in any genre. To do this, MusicLM has been trained on a data set of 280,000 hours of music.

The queries could be anything from "charming jazz song with catchy saxophone solo and solo singer", "90s Berlin techno with low bass and a strong beat", "basic soundtrack to an arcade game", for example. A neural network can even capture the desired mood and create music for a specific action or state: for meditation or awakening.

MusicLM can also "order" a specific musical instrument and even the level of training of the "musician" - for example, "novice pianist" and "virtuoso pianist".

But the AI is far from perfect - apparently, that's why Google doesn't want to make it publicly available and release the source code just yet. Some of the samples are of distorted quality, and voices and "lyrics" sometimes sound very bad. The developers have also discovered that around 1% of the music generated by the system was directly derived from songs that the neural network was trained on. And this could cause copyright issues.

You can listen to examples of music generated by MusicLM on GitHub.

Source: GitHub
Photo: Getty Images