SPECTRAL INTEGRATION AND NEURAL REPRESENTATION OF HARMONIC COMPLEX TONES IN PRIMATE AUDITORY CORTEX

Abstract

Many natural and man-made sounds, such as animal vocalizations, human speech, and sounds from many musical instruments contain rich harmonic structures. Although the peripheral auditory system decomposes these sounds into separate frequency channels, harmonically related frequency components must be grouped together in order to form a single auditory percept. A central neural process is therefore required to accomplish this perceptual grouping and to integrate information across frequency channels in order to compute spectral properties, such as pitch and timber, which are not explicitly encoded in the auditory periphery. In this dissertation, I investigated whether there are representations of harmonic structures at the single neuron level in auditory cortex beyond pitch and how harmonic sounds are represented by populations of cortical neurons. I systematically tested single neurons in the primary auditory cortex (A1) of awake marmoset monkeys with harmonic and inharmonic complex tones, varying fundamental frequency (f0) and harmonic composition. I found harmonic template neurons, which were strongly driven by harmonic complex tones but showed weak or no response to single harmonics. Harmonic template neurons were selective to f0s and sensitive to harmonic numbers. They also exhibited a reduced firing rate in response to inharmonic complex tones. Other sound features of a harmonic complex tone, such as overall sound level, resolved individual harmonic partials, and temporal envelope were represented by different subpopulations of neurons in A1. Overall, the findings of this dissertation support the existence of a distributed neural code for harmonic complex tones in A1 which represents an important stage in the auditory pathway for robust feature extraction and sound source recognition. In the study of spectral integration and neural coding of complex tones, searching for preferred stimuli of cortical neurons has also proven challenging because of the high dimensionality of the acoustic space of possible stimuli and limited recording time. In the last part of this dissertation, I presented an online adaptive stimulus design approach based on a neural network model for studying spectral integration in auditory cortex. The models estimated online helped to build a connection between receptive field structures and diverse spectral selectivity of cortical neurons.