Mechanisms of perceiving communication sounds in scenes

Abstract

Vocal communicators must perceive the vocal signals of social partners in complex auditory scenes that include distracting background sounds. The auditory system must therefore parse auditory scenes into multiple information streams and/or accurately encode individual vocalizations despite the presence of competing sounds. Explaining mechanisms whereby neural representations of vocalizations are extracted from neural representations of scenes is an important part of understanding how auditory processing leads to perception of communication signals in complex scenes. We study how songbirds recognize individual vocalizations (songs) in scenes of conspecific choruses. We combine behavioral studies with neurophysiological studies of song and scene coding in midbrain and cortex. We find dramatic transformations in the neural coding of songs and scenes between different regions of auditory cortex. Neural representations of individual songs are dense and non-selective in the midbrain and primary cortex, but are sparse and highly selective in higher cortex. Sparse coding neurons produce background-invariant responses to individual songs in scenes, providing a potential neural mechanism for the perception of individual communication vocalizations in complex auditory scenes. Acoustic manipulations of song and pharmacological manipulations of neural coding suggest that sparse and background-invariant representations of songs in higher cortex are due to context-dependent inhibition.