Rate Versus Temporal Code? A Spatio-Temporal Coherence Model of the Cortical Basis of Streaming

Abstract A better understanding of auditory scene analysis requires uncovering the brain processes that govern the segregation of sound patterns into perceptual streams. Existing models of auditory streaming emphasize tonotopic or “spatial” separation of neural responses as the primary determinant of stream segregation. While partially true, this theory is far from complete. It overlooks the involvement of and interaction between both “sequential” and “simultaneous” grouping mechanisms in the process of scene analysis. Here, we describe a new neuro-computational model of auditory streaming. Inspired by recent psychophysical (cf. abstract by Micheyl et al.) and physiological findings, this model is based on the premise that perceived segregation results from spatio-temporal incoherence, rather than just tonotopic separation. While tonotopic separation still plays an important role in this model, it is an indirect one: tonotopic overlap tends to reduce temporal incoherence, which in turn impedes segregation. The model simulates responses at the level of the primary auditory cortex and performs a correlative analysis of cortical responses in order to assess how different sound elements evolve in time in relation to each other. An eigenvector decomposition of this coherence analysis is used to predict how the input stimulus is organized into streams. The model is evaluated by comparing its neural and perceptual predictions under various stimulus conditions to physiological and psychophysical results.