Seoul National University researchers decoded relative pitch from electrodes on the brains of 10 epilepsy patients, reconstructing the shape of tunes they were only imagining.

Most people can play a song in their heads. A few notes of a familiar tune are often enough for the rest of the melody to unfold in the mind’s ear, complete with rises and falls. That inner music is private, but a new study suggests it leaves a trace that can be read from the brain.

Researchers at Seoul National University report in the journal eNeuro that they could reconstruct the contours of melodies people were imagining from recordings of their brain activity. The work was carried out by Jii Kwon and Chun Kee Chung.

Listening, imagining, humming

The study involved 10 patients with epilepsy who had electrodes placed on the surface of their brains as part of their medical care. Such recordings, known as electrocorticography, are used to locate the source of seizures before surgery, and they give researchers an unusually detailed view of brain activity.

Electrocorticography records electrical activity directly from the brain’s surface, beneath the skull. Compared with scalp EEG, it captures signals with much finer spatial and temporal detail, because the skull and scalp blur and weaken brain signals before they reach external electrodes. That precision makes it a valuable tool for studying how the brain represents complex information such as speech and music.

Participants heard the opening bars of children’s songs. They then imagined how each song continued and afterward hummed the melody they had imagined, giving the researchers a record of what was playing in their heads.

Do, re, mi

The team found that brain activity carried information about relative pitch, the relationships between notes rather than their exact frequencies. Their model was designed to decode those relationships.

“The model we created decodes relative pitch classes—such as do, re, mi, fa, sol, and la—rather than exact, absolute pitches,” Kwon said. “This is useful because people often recognize melodies by the relationships between notes, even when the same melody is played in a different key.”

That mirrors how most people experience music. Few can name the exact pitch of a note, an ability called absolute or perfect pitch. But almost anyone can recognize “Happy Birthday” whether it is sung by a child or a bass, because the pattern of steps up and down stays the same.

Relative pitch also helps explain why a melody remains recognizable when sung by different voices or played on different instruments. The brain appears to track the intervals between notes, whether a melody moves up by a small step or jumps by a larger leap, and to store tunes as patterns of these changes. Brain regions involved in processing pitch relationships overlap with those used for understanding the rises and falls of speech, which carry meaning in many languages.

By combining the model’s note-by-note predictions over time, the researchers reconstructed changes in melody that followed the structure of the songs participants imagined.

Part of a wider effort

The study adds to a growing body of research on decoding thoughts and perceptions from brain activity. Scientists have used similar approaches to reconstruct speech that people hear or intend to say, and in 2023 a team reconstructed a recognizable version of a Pink Floyd song from the brain activity of patients listening to it. Decoding imagined music is harder, because there is no external sound to align with brain signals.

Brain-computer interfaces that decode attempted speech have already allowed some people with paralysis to communicate. Kwon hopes the music research might eventually help people express what they are imagining, including those with little musical background and patients with conditions such as amyotrophic lateral sclerosis, or ALS, who may be unable to describe their inner experience.

Limits

The study was small, and electrocorticography requires surgery, so the approach is limited to patients who already need electrodes for medical reasons. The songs were familiar children’s tunes with simple structures, and it is not yet clear whether the method would work for unfamiliar or more complex music.

Imagined experiences also vary from person to person, and humming afterward offers only an approximate record of what someone heard in their head. Participants may have imagined the songs at different speeds or with slight variations in pitch, which could make decoding harder.

The authors suggest that future work could decode other features of music, such as tempo and dynamics, and test whether the approach extends beyond well-known songs. Noninvasive recording methods, which are more practical but less precise, would be needed for any broader use.

The work was funded by the Samsung Research Funding & Incubation Center of Samsung Electronics and Korea’s Ministry of Trade, Industry and Energy. It offers a glimpse of a future in which the tunes stuck in our heads might be heard by someone else.

---

Notes

1. Kwon, J., and Chung, C. K. “Reconstruction of Imagined Melody with Relative Pitch Decoding in Electrocorticography.” eNeuro, 2026. DOI: 10.1523/ENEURO.0289-25.2026.
2. Society for Neuroscience. “Brain activity reveals the melodies that people imagine.” EurekAlert!, 3 August 2026. https://www.eurekalert.org/news-releases/1137197.

Photo by Andrea Piacquadio on Pexels.com