Identification of natural and synthesized vowels produced by children and adults: Effects of fundamental frequency variation

Abstract

Fundamental frequency (F0) and formant frequencies (F1–F4) were measured for vowels from /hVd/ words produced by 10 men, 10 women, and 30 children (ages 3, 5, 7). For all age groups, intrinsic F (IF0) differences were present, i.e., high (open) vowels had higher F0’s than low vowels. However, there was increased F0 variability for younger children. Theoretically, the vocal tract transfer function is less precisely defined in the vowel spectrum when F0 is high. Therefore, time-varying changes in F0 could help delineate the shape of the transfer function. Also, IF0 might help specify vowel identity. To determine the perceptual consequences of F0 variation, natural and synthesized versions of a subset of the vowels were presented to listeners for identification. Although the natural vowels were identified more accurately than the synthesized vowels, there was no reduction in accuracy when the natural F0 contour was held constant over the time course of the vowel. Moreover, accuracy was not reduced when IF0 differences were eliminated by synthesizing each vowel with the speaker’s average F0. These results suggest that F0 variation does not play a prominent role in vowel identification for isolated syllables produced by children and adults. [Work supported by Texas Advanced Research Program.]