Effects of frequency scale on inter-age comparisons of first and second formant transitions

Abstract

Comparison of spectral measures from a wide age range of talkers, to draw inferences about inter-age differences in articulatory behavior, is confounded by the absolute resonant frequency differences that result from vocal tract size differences. Normalization procedures are required so that spectral differences due to vocal tract size differences are not misinterpreted as age differences in speech production behavior. Based on vowel (steady-state) data, the log Hz scale has been favored over several other frequency scales that have been used to normalize spectra for analysis of production data [T. M. Nearey, Proc. Int. Conf. Spok. Lang. Process. 1, 583–586 (1992)]. The magnitudes of first and second formant transition extents of male talkers, who ranged from infancy to adulthood, were compared under three different normalization conditions (log Hz, ERB, and Bark scales). Each formant was measured in Hz at the onset and offset of the consonant–vowel transition in wideband spectrographic displays of the syllables /bɑi /, /bi/, and /dæ/. These values were transformed to each scale and formant transition extents and rates were determined. The results are discussed relative to the potential of these scales to lead to artifactual differences in inferred inter-age articulatory differences.