Vowel normalization and ear-canal resonance

Abstract

One male and one female subject spoke words of the form/h_d/ containing each of the English vowels. The words were recorded simultaneously in two different conditions: (a) in the free field, and (b) through an ear of an acoustic manikin (“Kemar”) fitted with a Zwislocki coupler. The recordings were transferred to disk for analysis by the Speech Microscope [Vemula et al., J. Acoust. Soc. Am. 65, S22 (1979)]. Each vowel was sampled at midpoint, and comparison was made of the waveforms and spectra in both recording conditions. The results indicate that, in Kemar's ear, the spectrum envelopes of front vowels have approximately the same shape above 2000 Hz for both male and female talkers, despite the fact that the actual formant frequencies may be very different. As Kemar hears it, the vowel /i/ has a single peak centered at 3000 Hz, which may correspond to F2 for females and Fa for males. For back and low vowels, the central peak in the spectrum is higher for females than for males. These results suggest that high-formant (front) vowels are partially normalized before they are “heard.” [Supported by NINCDS Grant NS03856.]