The Relation between Vowel Recognition and Measures of Frequency Resolution

Abstract

The purpose of this study was to employ measures of frequency resolution obtained from individual subjects to predict each subject's vowel recognition performance. Input filter patterns at six test frequencies were obtained from normal-hearing and hearing-impaired subjects. These patterns were used to correlate frequency resolution with vowel recognition in those same subjects. Vowels were presented at levels at which the entire spectrum was fully audible to each subject. Using each subject's measured filter characteristics (and interpolated values for intermediate frequencies), an "internal spectrum" of each vowel was calculated by determining the outputs of all filter channels for the vowel as the input signal. It was speculated that the more similar two internal spectra for a subject were, the more often they would be confused in the vowel recognition task. This expectation received some support when the measure of similarity was a point-by-point Euclidean distance between the two internal spectra. Stronger support was obtained when the measure of similarity was based upon Klatt's (1982) "weighted slope metric" that emphasizes similarities of spectral peak locations. The present study demonstrates a relation between impairments of frequency resolution and vowel recognition. The described filter-bank model of vowel recognition suggests that measures of frequency resolution along with the acoustic spectra of vowel stimuli may be useful in predicting the recognition of vowels by individuals.