Schedule of Events | Symposia

Postdoctorial Fellowship Award Winner

Decoding speech and music from intracranial recordings: evidence for domain-general representations of sound in the human brain

Poster Session F - Tuesday, April 1, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Jérémie Ginzburg^1,2 (jeremie.ginzburg@mcgill.ca), Émilie Cloutier Debaque², Arthur Borderie², Laurence Martineau³, Paule Lessard Bonaventure³, Robert J Zatorre¹, Philippe Albouy²; ¹Montreal Neurological Institute, McGill University, Montreal, Canada, ²CERVO research center, Laval University, Québec, Canada, ³Centre Hospitalier Universitaire, Laval University, Québec, Canada

Speech and music represent the most complex ways through which humans use sound to convey information. Recent studies increasingly indicate that the human brain is finely tuned to encode the temporal and spectral features of sound that are crucial for processing speech and music, respectively. However, a key debate persists regarding the brain's representation of perceptual objects: does it rely on category-based encoding (domain-specific view) or physical sound features (domain-general view)? To investigate this question, we conducted stereo-electroencephalography recordings in eleven epilepsy patients with electrodes implanted into the auditory cortex as they listened to a natural stimulus (the audio track of a famous movie) containing both speech and music. Using multivariate ridge and logistic regression analyses, we decoded the spectrotemporal features of the sound from the brain activity. These decoded features were then used to predict the speech/music categories of the sound, based on independent behavioral judgments obtained from 19 healthy listeners. Separately, we decoded the spectrotemporal features of the audio signal underlying these perceptual categorizations. Strikingly high ranked correlations (0.77 < ρ < 0.91, all corrected p-values < .001) were observed between the spectrotemporal features decoded from brain activity that were used to predict perceptual categorization and the spectrotemporal features associated with perceptual categorization. These findings suggest that the categorization of speech and music relies primarily on the spectrotemporal features of sound, providing robust evidence in favor of the domain-general view that higher-order sound representations in the brain are grounded in their physical properties.

Topic Area: PERCEPTION & ACTION: Audition