SALIENCE listener vs EEG: results (Bach dataset)
Year: 2026
Type: preprint
Abstract: Pre-registered test (plan: doi:10.5281/zenodo.23961990) of whether the surprise computed by the SALIENCE model listener predicts human EEG during music listening beyond acoustic envelope and spectral flux. Open data from Di Liberto et al. (2020): 20 listeners, monophonic Bach melodies. Adding the listener’s surprise improved held-out EEG prediction in 17 of 20 listeners (p = 0.0003, Holm-corrected; d = 0.82). Surprise computed from a mismatched piece added nothing. The result held with a second piano rendering, with the original audio envelope in the baseline, and beyond score-based melodic expectation (IDyOM). The effect is small in absolute terms (about 4 % more prediction accuracy). Replication on produced electronic music (NMED-T) is pending. Includes analysis code (MIT), features, per-participant scores, null distributions and a dated deviation log.
Author: Psykovsky, Vasily
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7225254989