Acoustic comparison of unvoiced fricatives in telephone speech versus typical speech

Abstract

The results of a pilot study designed to compare the acoustic properties of unvoiced fricatives in telephone speech versus speech typically used for acoustic‐phonetic analyses are reported. Two new characterizing parameters for the unvoiced fricatives are introduced: the half‐power bandwidth (resonance width) and a three‐point energy versus time contour. It is found that the average fricative power normalized by the average adjacent vowel power is considerably larger in the telephone speech for all speakers, fricatives, and fricative positions studied. Also, half‐power bandwidths tend to be larger for [f] and [θ] in telephone speech, but smaller for [s] and [∫]. The study yields a preliminary understanding of these acoustic differences in terms of the differences in vocal effort, microphone placement, and frequency response and bandwidth of the recording equipment. Finally, possible distinguishing perceptual cues for the different recording environments in terms of the acoustic differences are hypothesized. In particular, the energy‐time contour of [θ] exhibits a peculiar high‐low‐high pattern that is most pronounced in the telephone recordings. [Work supported by AT&T, NSF, and the University of Kansas.]