Researchers from Google joined Japan’s Osaka University to conduct a study that saw them convert brain activity into streamable music.
For the effort, five volunteers underwent fMRI scans while being played music from an assortment of 500 tracks across 10 genres.
Researchers then took the scans of the volunteers’ brain activity and fed them into an AI model they had built called Brain2Music. The program draws information from the scans to produce songs inspired by those the volunteers were listening to.
The new AI-generated songs came from Google’s production tool, MusicLM, which is a text-to-music model that churns out tracks based on text descriptions.
In the published study, the researchers described the AI music as resembling “the musical stimuli that human subjects experienced, with respect to semantic properties like genre, instrumentation, and mood.”
They also wrote about how the exercise allows them to study the “brain regions represent information derived from purely textual descriptions of music stimuli.”
While conducting an fMRI definitely isn’t the most convenient way to make music, the study offers insight into the relationship between brain activity and the stimulus that music provides, as well as a look at the impressive capabilities of Google’s AI.
In other music news, a 1976 Apple check Signed by Steve Jobs and Steve Wozniak is up for auction.
