AI just recreated Pink Floyd straight from human brainwaves

Scientists just found a way to pull rock music straight out of human brainwaves.

Researchers at the University of California, Berkeley, had 29 epilepsy patients listen to Pink Floyd’s classic track “Another Brick in the Wall, Part 1” while recording their intracranial brain activity. An AI model then translated those neural patterns back into a recognizable, acoustic audio stream.

It is the first time researchers have successfully decoded a complex song from brain recordings, opening up new possibilities for both neurotechnology and human communication.

Reverse-engineering Pink Floyd

The patients involved in the study already had electrode grids surgically implanted on the surface of their brains to treat epilepsy.

As the music played, the team mapped which electrodes lit up in response to specific musical elements like pitch, melody, harmony, and rhythm. They identified a specific region inside the superior temporal gyrus that processes the rhythm of the guitar, and confirmed that the right hemisphere does most of the heavy lifting when it comes to musical processing.

To test the AI, the researchers held back a 15-second slice of the song during training. When fed only the brain signals from that hidden window, the AI generated an audio spectrogram that was 43 percent similar to the original clip.

The resulting audio sounds a bit like Pink Floyd playing underwater, but the cadence, guitar strums, and vocal melody are unmistakably intact.

Whose song is it anyway?

The team believes this could eventually work in reverse. Study co-author Ludovic Bellier noted that if neural decoding can reconstruct what you hear, it might soon decode what you imagine, effectively letting people compose music with pure thought.

Naturally, that is already raising eyebrows in the legal world.

Jennifer Maisel, an attorney at Rothwell Figg, pointed out that reconstructing recognizable tracks from brain activity will quickly run into copyright disputes. The authorship question is equally murky: who owns a song decoded from brainwaves? The patient, the recording technician, or the AI developer?

Linklaters attorney Ceyhun Pehlivan added that ownership might even hinge on neuroanatomy. A signal pulled from the sensory auditory cortex looks like a passive recording, whereas signals decoded from the frontal cortex could represent creative human intent.

Why it matters

This is not about building a brain-controlled Spotify player for everyday users. The real breakthrough here belongs to assistive technology.

Current speech neuroprosthetics can help people with conditions like ALS or aphasia produce text, but the resulting synthesized speech sounds entirely robotic. Human speech relies heavily on musical elements—prosody, emotional inflection, and rhythm. By figuring out how the brain encodes those elements, researchers can finally build speech devices that sound like an actual person having a conversation.

Don't expect consumer headsets doing this anytime soon. Lead researcher Robert Knight made it clear that invasive brain implants will remain strictly clinical. While other teams are experimenting with non-invasive MRI scans, their spatial and temporal resolution still lags far behind.

We think Berkeley’s clever move was choosing music rather than spoken words. Music pushes acoustic dynamic range to the extreme; if an AI can reconstruct rock guitars and vocal harmonies from cortical signals, decoding the emotional tone of everyday speech suddenly looks much closer to reality.