Adaptive estimation of phoneme boundaries and selective adaptation for speech.

Abstract

Previous studies have estimated the boundary between two phoneme categories from identification data. This procedure is time consuming, especially in studies of selective adaptation, in which it is necessary to test identification twice, before and during adaptation. We adapted a simple up-down adaptive procedure to estimate the phoneme boundary. The stimulus to be presented on a given trial is determined from the subjects’ past responses. Two stimulus continua were synthesized: a voicing continuum (13 stimuli changing from /ba/ to /pa/) and a place continuum (13 stimuli changing from /ba/ to /da /and /ga/). Twenty-two normal listeners were given an identification test with each set of stimuli. A logistic curve was fit to the obtained identification function from each subject. Next, phoneme boundaries were directly estimated with the adaptive procedure. The 50% points on the curves from the identification procedure were highly correlated with those estimated by the adaptive procedure. Phoneme boundaries were also measured before and during selective adaptation fusing all end-point stimuli as adaptors in separate sessions). Boundary shifts were obtained in the expected directions and magnitudes. The adaptive procedure is faster than the identification test and exhibits high test-retest reliability.