Correlation-based temporal model of pitch multiplicity

Abstract

Global temporal models for pitch analyze population-wide, all-order interspike interval distributions of the entire auditory nerve to accurately predict almost all known F0-pitch phenomena. Population-interval distributions (PIDs) are neural representations that resemble log-frequency scaled, half-wave rectified summary autocorrelation functions (SACF) of stimuli. PIDs produced by periodic stimuli show regular patterns of lag peaks associated with subharmonics of harmonics, with major peaks at n/F0, whereas early global temporal models chose the highest PID peak to estimate one, dominant pitch, later models compared average interval-densities of different sets of F0-related PID peaks to estimate relative pitch saliences, all saliences above a threshold being audible. However, this method inherently generates octave confusions amongst related subharmonics. Using Pearson correlation coefficients between PIDs and periodicity-related lag patterns (lags at n/F0) to estimate pitch saliences obviates the octave ambiguity problem and lag-weightings. The results from recent simulations using the Zilany-Bruce-Carney ANF model to estimate the pitches heard for dyads and triads of complex harmonic tones will be presented. For triadic chords (major, minor, suspended, augmented, and diminished), the model estimates relative saliences of F0s of notes, fundamental basses, and individual harmonics). Correlation salience values indicating relative pitch strengths (pitch stabilities) generally comport with music theoretic rankings for these chords.