Overview
- Researchers at Seoul National University led by Jii Kwon and Chun Kee Chung used electrocorticography from ten epilepsy patients to record brain activity while participants listened to a few bars, silently imagined the continuation, and then hummed the melody.
- The team trained a decoder that targets relative pitch classes—do, re, mi and so on—rather than exact frequencies and then combined note-level predictions over time to reconstruct the melodic contour of the imagined tunes.
- The reconstructed melodies reflected the sequence of imagined notes well enough to show that internal musical representations leave a measurable trace in invasive cortical signals.
- The study is a proof of concept published in eNeuro, but it is limited by a small clinical sample, use of familiar children’s songs, and the focus on pitch only rather than tempo, dynamics, or broader repertoires.
- If validated in larger and noninvasive samples, the approach could inform music-based brain‑computer interfaces for nonverbal patients, but substantial technical, ethical, and clinical translation challenges remain.