Scientists Decode the Songs Playing in Your Head

Researchers have taken a striking step toward reading the music inside people’s minds. A team at Seoul National University recorded brain signals from epilepsy patients as they imagined familiar tunes. The work shows that patterns in those signals can reveal the relative pitches of notes the subjects never sang aloud.

The study, published in eNeuro, involved 10 patients who already had electrodes placed on the surfaces of their brains for clinical monitoring. Participants listened to the opening bars of children’s songs. They then silently imagined how the melody continued before humming it out loud. Electrodes captured high-gamma activity and other frequency bands during that quiet imagery phase.

Jii Kwon, the lead author, explained the approach. “The model we created decodes relative pitch classes-such as do, re, mi, fa, sol, and la-rather than exact, absolute pitches. This is useful because people often recognize melodies by the relationships between notes, even when the same melody is played in a different key.” The focus on relative pitch proved decisive. Models trained on absolute pitch performed at chance levels. Yet relative-pitch classifiers exceeded chance, even if single-note accuracy remained modest at about 27 percent on held-out test data.

That modest per-note success still delivered something remarkable. Researchers chained the predicted pitch classes together across time. The resulting sequences preserved the overall contour of the imagined melodies. Spearman correlations between reconstructed and actual melodic shapes averaged 0.67, statistically significant. Ups and downs in pitch emerged clearly even when individual notes occasionally strayed. The brain, it turns out, broadcasts enough structure during internal rehearsal to let outsiders trace the tune.

Activity in the superior temporal gyrus, especially on the left, carried the strongest signals. Precentral regions, tied to motor planning, contributed as well through beta and delta bands. These findings align with earlier noninvasive work. A 2024 study in PLOS Biology by David R. Quiroga-Martinez and colleagues used magnetoencephalography on 71 volunteers. They decoded both heard and imagined melodies. Auditory cortex handled the sensory details of individual sounds during listening. A broader network spanning frontal and parietal areas, hippocampus, basal ganglia and sensorimotor cortex represented the melody as an abstract sequence in both perception and imagination.

That paper highlighted volitional control. When participants mentally manipulated a melody, the neural pattern shifted systematically. “We found that, during perception, auditory regions represent the sensory properties of individual sounds,” the authors wrote. “In contrast, a widespread network including fronto-parietal cortex, hippocampus, basal nuclei, and sensorimotor regions hold the melody as an abstract unit during both perception and imagination.” The mental act of changing the image left a detectable fingerprint. Such results suggest imagination isn’t mere replay. It engages executive and memory systems that reshape the representation on the fly.

But the new intracranial data go further. They offer higher spatial and temporal precision. Surface electrodes sit closer to the cortex than scalp EEG. They pick up local field potentials with less distortion. The patients performed the task with familiar children’s songs presented in multiple keys. That design forced the brain to rely on relational knowledge rather than absolute frequency memory. And the electrodes revealed exactly that. Notes sharing the same absolute pitch but different scale degrees produced distinguishable signals. The brain encodes musical meaning, not just raw sound.

Earlier efforts had hinted at this possibility. A 2021 study decoded imagined Bach melodies from professional musicians’ EEG with notable accuracy at the individual-trial level. Another team reconstructed a Pink Floyd song from brain activity during listening. Yet reconstructing purely imagined sequences without any auditory playback remained harder. The Seoul team succeeded by abandoning the quest for perfect note-by-note fidelity. They bet on sequence instead. The gamble paid off. Reconstructed contours matched the target melodies well enough to trace the tune’s shape.

And the implications stretch beyond music. Patients with amyotrophic lateral sclerosis often lose speech while retaining rich inner lives. Some may still imagine music vividly. If future systems can decode those internal melodies, they might offer a novel communication channel. Kwon and senior author Chun Kee Chung see that potential. They suggest the technique could extend to people with little musical training. It might one day capture tempo, dynamics or even emotional tone. But those steps remain distant. Current models work best on familiar material. Generalizing to novel compositions or untrained subjects will demand larger datasets and smarter algorithms.

The work also sharpens the picture of how the brain simulates sound. Predictive coding theories propose that auditory cortex constantly forecasts upcoming input and compares it against reality. During imagery the forecast runs without external input. The same error signals that guide perception may sculpt the internal experience. Earlier research from the University of Maryland showed that silences embedded in music trigger brain responses similar to those during active imagination. The brain fills in the missing notes. It does so using the same machinery that processes real sound.

Yet important differences exist. Noninvasive MEG studies show imagination relies more heavily on association areas and less on primary auditory cortex than listening does. The intracranial recordings reinforce that view. High-gamma bursts in superior temporal gyrus still appear, but motor-related signals in precentral cortex gain prominence. Imagination, after all, often feels like a faint echo of singing or playing. The body stays still. The motor plan still fires.

Critics may worry about privacy. If machines can pull melodies from silent thought, what else might they extract? The current technology requires implanted electrodes. It works only in controlled lab settings with cooperative patients. Scalp-based systems lag far behind in precision. Widespread mind-reading of casual earworms sits years away. Still, the pace of progress in brain-computer interfaces has accelerated. Speech-decoding implants have already restored communication for paralyzed individuals. Music could follow.

Recent coverage highlights the excitement. Neuroscience News reported the findings under the headline that songs imagined inside the head can now be decoded directly from brain signals. The article stressed the relative-pitch insight and its value for future assistive devices. Similar pieces in medical outlets echoed the hope that such methods might help patients who cannot speak or move but who retain musical memory.

So what comes next? Larger cohorts. Noninvasive alternatives that approach the fidelity of intracranial recordings. Experiments with untrained listeners and unfamiliar music. Integration with generative audio models that could synthesize the decoded melody in a chosen instrument or voice. The technical barriers are real. Single-note accuracy must improve. Cross-subject generalization needs refinement. Yet the proof of concept stands. The brain’s internal concert leaves traces. Scientists have begun to transcribe them.

That transcription remains imperfect. Reconstructed melodies sometimes wander. Contours bend in the wrong places. But the core shape survives. Listeners can often recognize the tune. The gap between mind and machine narrows. Not because the brain has grown simpler. Because the tools have grown sharper. And the questions have grown bolder.

Researchers once asked whether imagination and perception share neural codes. They do. Now they ask whether those codes can be read in real time. The answer, at least for relative pitch in familiar songs, is yes. The next movement in this scientific score will test how far the orchestra of electrodes and algorithms can follow the silent symphony inside each of us.