A New Approach to Model Pitch Perception Using Sparse Coding

Oded Barzelay, Miriam Furst, Omri Barak

Abstract: Author Summary By means of a sound's pitch, we can easily discern between low and high musical notes, regardless of whether they originate from a guitar, piano or a vocalist. The relation between different sounds that yield the same percept is what makes pitch an interesting subject of research. Today, despite extensive research, the mechanism behind this physical to perceptual transformation is still unclear. The large dynamic range of the cochlea combined with its nonlinear nature makes the modeling and understanding of this process a challenging task. Given a large amount of physiological and psychological data, a general explanation consistent with many of these phenomena would be a major step in elucidating the nature of pitch perception. In this paper, we recast the problem in the general framework of sparse coding of sensory stimuli. This framework, initially developed for the visual modality, posits that the goal of the neural representation is to represent the flow of sensory information in a concise and parsimonious way. We show that applying this principle to the problem of pitch perception can explain many perceptual phenomena.