Brain2Music : an AI that generates music from your thoughts !

Google showed off a new AI that can create music by interpreting your brain activity.

Imagine a world where your thoughts could compose symphonies, where the melodies dancing through your mind find their way into reality. That vision is now closer to realization, thanks to the groundbreaking collaboration between Google, Osaka University, the National Institute of Information and Communications Technology, and Araya. Enter Brain2Music, Google's latest AI marvel that transcends conventional musical creation by harnessing the enigmatic power of your brain activity.

In a study called “Brain2music : Reconstructing Music from Human Brain Activity,” researchers from Osaka University and the National Institute of Information and Communications Technology in collaboration with Google and Araya presented the fruit of their work on Artificial Intelligence in relation to human cerebral activity.

They created a “method of reconstructing music from brain activity captured using functional magnetic resonance imaging.” To do this, the scientists used MusicLM, an AI capable of generating music from a text description designed by Google. Rather than creating music from text, Brain2Music's AI interprets brain activity data obtained from MRI scans.

BRAIN2MUSIC IS GOOGLE'S NEW AI THAT TURNS YOUR THOUGHTS TO MUSIC

To train the neural networks behind Brain2Music, the research team subjected five subjects to 15-second samples of music in styles as varied as rock, reggae, jazz or metal. The videos obtained through functional magnetic resonance imaging then allowed the AI to make the connection between the patterns and the areas of the brain activated according to the different characteristics of the music heard (rhythm, mood, dynamics, for example ). Once this data had been interpreted and translated into words, all that remained was to provide it to MusicLM, so that it could “recreate” the original piece.

According to the working group, the results of the experiment are very conclusive : "the generated music resembles musical stimuli that human subjects have experienced with respect to semantic properties such as genre, instrumentation and mood" . Composers who wish to create music “in thought” will still have to wait a while. If the AI ​​achieves a great feat, the reconstructed pieces are far from being faithful to the source.