Auditory-inspired pitch extraction using a synchrony capture filterbank for speech signals

Abstract

The question of how harmonic sounds produce strong, low pitches at their fundamental frequencies, F0s, has been of theoretical and practical interest to scientists and engineers for many decades. Currently the best auditory models for F0 pitch, [e.g., Meddis and Hewitt, J. Acoust. Soc. Am. 89(6), 2866–2894 (1991)] are based on bandpass filtering (cochlear mechanics), half-wave rectification and low-pass filtering (haircell transduction and synaptic transmission), channel autocorrelations (all-order interspike interval statistics) aggregated into a summary autocorrelation, and an analysis that determines the most prevalent interspike intervals. As a possible alternative to autocorrelation computations, we propose an alternative model that uses an adaptive Synchrony Capture Filterbank (SCFB) in which groups of filter channels in a spectral neighborhood are driven exclusively (captured) by dominant frequency components that are closest to them. The channel outputs (for frequencies below 1500 Hz) are then adaptively phase aligned with respect to a common time reference to compute a Summary Phase Aligned Function (SPAF), aggregated across all channels, from which F0 is easily extracted.