Prediction and Evaluation of Vocal Performance in the Songbird Auditory Cortex

Abstract

Learning and maintaining complex vocalizations requires the brain to generate avocal motor program that matches an auditory target. One possibility is that this matching process depends on neural circuits that compare vocal auditory feedback to an internal representation of the auditory target, and differences in these two representations result in an error signal that is used to modify the vocal motor program. This comparison is thought to be facilitated by vocal-motor or corollary discharge signals that suppress auditory cortical responses to predictable features of vocalization-related auditory feedback. Yet whether predictive suppression characterizes the auditory cortex in animals that produce learned vocalizations remains poorly understood. I explored this issue in freely behaving adult male zebra finches, which use auditory feedback to learn and maintain their courtship songs. I used miniature microscopes and genetically encoded calcium indicators to measure how vocal-motor signals modulate auditory cortical activity. I found many auditory cortical neurons that were suppressed during singing, even though these same neurons were excited during non-singing epochs by playback of the bird's own song. Additionally, a small proportion of neurons exhibited suppressed activity several hundred milliseconds prior to song onset, consistent with a vocal corollary discharge signal. Finally, perturbing auditory feedback with singing-triggered noise also revealed a subpopulation of neurons that behaved like error-detectors, consistent with a predictive coding process. To further explore the extent to which the singing-related activity of auditory cortical neurons depend on auditory feedback, I tracked activity before and after deafening, while also using a variational autoencoder to measure deafening-induced changes to song. I found auditory cortical neurons that maintained similar patterns of singing-related modulation following deafening, but these activity patterns became more variable in parallel with changes in song structure. Other neurons that were suppressed or inactive during singing prior to deafening became strongly excited during singing after deafening, supporting the hypothesis of an emergent error-like signal that drives changes in the song. Altogether, these results provide evidence of a vocal corollary discharge signal that functions predictively in the auditory cortex to suppress singing-related feedback and suggest that this signal contributes to the maintenance of stable vocal motor programs.