Prediction of listener perception of reduced, voice stop consonant simulations based on patterns of formant deflections

Abstract

Previous research on stop consonants found that less than 60 percent of the stops sampled from a speech corpus contained a clearly defined period of silence or prevoicing prior to the plosive release [Crystal & House, JASA, 1988]. How listeners perceive a reduced form of stop consonants without these cues is not well understood. The purpose of this experiment was to investigate whether recasting typical formant transitions into a measure called a “relative formant deflection pattern” provides a means of predicting listeners’ perceptions of approximant-like, voiced stop consonant variants. A computational model of speech production, in which consonant constriction location was varied along the length of the vocal tract, was used to generate place continua of approximate-like, voiced stop consonants imposed on a vowel-to-vowel transition. Stimuli were presented to listeners in three conditions: 1) normal simulated speech, 2) sinewave speech in which three tones replicated the time course of the F1, F2, and F3 contours in the simulated samples, and 3) sinewave speech in which three tones were present, but selected combinations of F1, F2, and F3 were set to a flat contour. Perceptual responses will be compared to the predictions based on relative formant deflection patterns across conditions.