3aSC4. A model of silence duration adaptation in the perception of consonant cluster pairs.

Session: Wednesday Morning, June 18

Author: Michael A. Cohen
Location: Cognit. and Neural Systems Dept., Boston Univ., Boston, MA 02215
Author: Krishna K. Govindarajan
Location: MIT, Cambridge, MA 02139


Listeners are able to adapt to the silence duration distribution between two different stop consonants VC[inf 1]-C[inf 2]V [K. K. Govindarajan and M. A. Cohen, J. Acoust. Soc. Am. 95, 2978(A) (1994)]. This adaptation affects the likelihood of hearing just one stop consonant or the consonant cluster. An adaptation model is presented that emulates most of the individual subject data reported earlier; and based on Chi square goodness-of-fit tests, is capable of accounting for 98.5% of the variance of the psychometric functions. The model consists of a one-stage leaky integrator that averages the silence duration of prior tokens. This average is subtracted from the current silence duration which is then corrupted by Gaussian noise of constant mean and variance. If the result is positive, the predicted response is hearing two stops, otherwise one stop is heard. The decay rate, the initial average, and the mean and variance of the noise were fit to the six subjects in a variety of experiments and conditions using maximum likelihood estimation. Reasons for the reliability of the constructed fit, and variability of parameters across subjects and conditions will be discussed. [Work supported by AFOSR, NIH.]

ASA 133rd meeting - Penn State, June 1997