Perceived phonetic distance among a set of synthetic whispered vowels and fricative consonants

Abstract

Phonetically trained listeners were presented with pairs of 300‐ms synthetic speech‐like sounds and asked to estimate phonetic distance between members of a pair. The first sound, the reference, was always the same—a voiceless [ae]. A set of 33 other sounds were synthesized by making small changes to formant frequencies, bandwidths, or spectral tile of the reference. Results indicate that phonetic distance judgements are consistent with previous work on voiced vowels [D. H. Klatt, J. Acoust. Soc. Am. 66, S86 (1979)] in showing that small formant frequency changes caused the largest perceived phonetic distances. However, the new results differ in that (1) changes in first formant frequency had little effect (probably because first formant bandwidth increases associated with an open glottis cause a relatively indistinct first formant spectral peak), and (2) spectral tilt changes also influenced distance judgments (probably by turning some stimuli into fricative‐like spectra). In order to check this latter hypothesis, the tape was played to a new group of subjects who were asked to indicate whether each stimulus was a vowel or a fricative. It was found that fricatives are heard when spectral tilt increases, or when third or fourth formant frequency changes result in a high‐frequency energy concentration. [Work supported in part by an NIH grant.]