A psychoacoustic model of the ba/wa boundary shift

Abstract

Previous studies have shown that the ba/wa boundary (cued by formant transition duration) shifts in the context of different vowel durations. The shift has the appearance of a “rate normalization,” but modeling of potential auditory factors involved in perception of transition durations suggests that the effect may be qualitatively different from other cases of rate normalization. The present model suggests that the shift of boundary may result from a difficulty of the perceiver to recognize exactly when a change in frequency trajectory (a pivot point) of a formant occurs. The model predicts an auditory overestimation of formant transition durations in cases where the following vowel is short, and a progressive correction of that overestimation as the vowel increases in duration. The model accounts for the magnitude of the ba/wa shift in speech and analogous nonspeech stimuli, and also provides a natural account of the asymptotic characteristic of the ba/wa shift (i.e., the fact that the shift essentially disappears after vowels reach 200 ms or more). Finally, the model offers a natural account of the previously unexplained fact that the ba/wa boundary sometimes shifts at times that the rate normalization hypothesis would not appear to predict (as in the case of lengthening syllables by adding arresting transitions such as occur in bad/wad).