The neural representation of song in human auditory cortex

University Of Rochester, Rochester NY

Investigators

Abstract

Song is present in every known human culture, fosters social bonding, and has substantial therapeutic value. To perceive song, the human brain must solve some of the most difficult challenges of hearing, including isolating singing from musical accompaniment and integrating speech, rhythmic, and melodic information. Recently, a neural population was discovered in the human brain that responds strongly to singing and weakly to both speech and instrumental music. The discovery of this neural population provides a unique opportunity to understand the brain mechanisms of song perception and how these mechanisms differ from those used to perceive other sounds including music and speech more generally. Leveraging this discovery, this project answers key questions about how song is coded in the human brain leveraging advanced experimental and computational methods. The project engages high school and undergraduate students from diverse backgrounds to participate in interdisciplinary research at the intersection of artificial intelligence, statistics, and neuroscience. The data and methods developed in this project are carefully documented and shared, ensuring they have a broad impact, and the project disseminates findings to a wide audience through, for example, science media and museums to help foster continued interest and support for scientific research. The project measures cortical responses to singing using functional MRI (fMRI) and intracranial recordings from human neurosurgical patients. Intracranial recordings provide a rare opportunity to measure human brain responses with very high precision, while fMRI studies make it possible to characterize neural responses across many different sounds and participants. The project develops improved statistical methods to isolate neural populations selective for singing and differentiate them from nearby neural populations that respond selectively to speech and music. The project uses computational audio methods to manipulate the melodic, rhythmic, and speech content of singing and then fit “encoding models” to determine how these different features of song are represented in different neural populations. Through these techniques, the project fundamentally improves the understanding of how human cortex encodes and processes song as a unique auditory stimulus. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.