Google’s New AI System Can Generate Music From Text Descriptions

google musiclm

MusicLM is an AI System developed by Google which can generate music of any genre from just a text description.

While this isn’t the first artificial intelligence system in this space, it definitely is the first which has overcome all the limitations of the others to actually do the job.

While Google is not planning to release this at the moment, a detailed academic paper revealed that MusicLM was trained on a dataset of 280,000 hours of music in order to learn how to generate coherent songs from what the creators explain as descriptions of significant complexity. Examples of some of those text descriptions include statements like, “Berlin ’90s techno with a low bass and strong kick.”

After the immense amount of training the system apparently generates songs that sound remarkably like something that’s composed by a human.

As reported by TechCrunch, MusicLM’s artificial intelligence capabilities extend beyond generating short clips of songs. The Google researchers show that the system can build on existing melodies, whether hummed, sung, whistled or played on an instrument. Moreover, MusicLM can take several descriptions written in sequence (e.g. “time to meditate,” “time to wake up,” “time to run,” “time to give 100%”) and create a sort of melodic “story” or narrative ranging up to several minutes in length — perfectly fit for a movie soundtrack.

Addressing why there aren’t any plans of releasing this publicly yet, the co-authors found during testing that about 1% of the music the system generated was directly replicated from the songs on which it trained and they wrote, “We acknowledge the risk of potential misappropriation of creative content associated to the use case. We strongly emphasize the need for more future work in tackling these risks associated to music generation.”

Read the entire paper here and listen to samples on Tech Crunch here!

Melody Siganporia
Follow me


Please enter your comment!
Please enter your name here