Multivoxel codes for representing and integrating acoustic features in human cortex

Abstract

Using fMRI and multivariate pattern analysis, we determined whether acoustic features are represented by independent or integrated neural codes in human cortex. Male and female listeners heard band-pass noise varying simultaneously in spectral (frequency) and temporal (amplitude-modulation [AM] rate) features. In the superior temporal plane, changes in multivoxel activity due to frequency were largely invariant with respect to AM rate (and vice versa), consistent with an independent representation. In contrast, in posterior parietal cortex, neural representation was exclusively integrated and tuned to specific conjunctions of frequency and AM features. Direct between-region comparisons show that whereas independent coding of frequency and AM weakened with increasing levels of the hierarchy, integrated coding strengthened at the transition between non-core and parietal cortex. Our findings support the notion that primary auditory cortex can represent component acoustic features in an independent fashion and suggest a role for parietal cortex in feature integration and the structuring of acoustic input.\n\nSignificance statementA major goal for neuroscience is discovering the sensory features to which the brain is tuned and how those features are integrated into cohesive perception. We used whole-brain human fMRI and a statistical modeling approach to quantify the extent to which sound features are represented separately or in an integrated fashion in cortical activity patterns. We show that frequency and AM rate, two acoustic features that are fundamental to characterizing biological important sounds such as speech, are represented separately in primary auditory cortex but in an integrated fashion in parietal cortex. These findings suggest that representations in primary auditory cortex can be simpler than previously thought and also implicate a role for parietal cortex in integrating features for coherent perception.