The zebra finch auditory cortex reconstructs occluded syllables in conspecific song

Abstract

Sensory input provides incomplete and often misleading information about the physical world. To compensate, the brain uses internal models to predict what the inputs should be from context, experience, and innate biases (Liberman and Mattingly, 1985; Komatsu, 2006; Gilbert and Sigman, 2007; Heald and Nusbaum, 2014). For example, when speech is interrupted by noise, humans perceive the missing sounds behind the noise (Miller and Licklider, 1950; Warren, 1970; Samuel, 1996), a perceptual illusion known as phonemic (or auditory) restoration. The neural mechanisms allowing the auditory system to generate predictions that override ascending sensory information remain poorly understood. Here, we show that the zebra finch (Taeniopygia guttata) exhibits auditory restoration of conspecific song both in a behavioral task and in neural recordings from the equivalent of auditory cortex. Decoding the responses of a population of single units to occluded songs reveals the spectrotemporal structure of the missing syllables. Surprisingly, restoration occurs under anesthesia and for songs that the bird has not heard. These results show that an internal model of the general structure of conspecific vocalizations can bias sensory processing without attention.