Modeling formant frequency discrimination for isolated English vowels using excitation patterns

Abstract

Thresholds for formant discrimination across three sets of female and male vowels with different F0 were significantly different in a recent report [Kewley-Port etal., J. Acoust. Soc. Am. 95, 2978(A) (1994)]. This analysis examined whether excitation patterns could model these and other effects of stimulus parameters on formant thresholds. The goal was to determine if an ‘‘auditory metric’’ would be constant across the three stimulus sets when ΔF thresholds varied by 25 Hz. A separate discrimination study showed that listeners only attend to harmonic components within a restricted region near the formant [Sommers and Kewley-Port, J. Acoust. Soc. Am. 93, 2422(A) (1993)]. Based on those results, four critical bands around the altered formant were selected, and the area between the critical-band spectra for the standard and just discriminable vowel was calculated. This spectral distance across formant frequency and gender was shown to be constant in three analyses: (1) ΔF threshold differences across the three sets of vowels were no longer significant; (2) slopes for ΔF thresholds (approximately 1.0) were flat for spectral distance; (3) variability of spectral distance across F1 and F2 is significantly smaller than that of ΔF thresholds. Results suggest that the auditory system has an inherent nonlinear transformation which changes threshold differences to be almost constant in the internal representation.