Vocal tract normalization for /s/ and /š/

Abstract

The acoustic cue separating /s/ and /š/ was studied, using synthetic speech stimuli, in the context of two different following [ae]'s: one appropriate for a large vocal tract and one for a small vocal tract. For each vocal tract size, the center frequency of the frictional portion of the stimuli was varied in ten steps from 2974 to 4455 Hz. In preliminary experiments, it was found that listeners have a reasonably consistent boundary along this scale at which they stop hearing /š/ and start hearing /s/. It was hypothesized that a listener's boundary would be higher for the small-vocal-tract stimuli than for the large-vocal-tract stimuli. Nineteen subjects who were asked to identify 40 tokens of each stimulus showed an upward boundary shift of varying magnitude for the small-vocal-tract stimuli as compared with the large-vocal-tract stimuli. This result suggests the /s/-/š/ cue is dependent on vocal tract size.