Listeners adapt their perception of speech to handle acoustic variability caused by factors such as dialects, accents, and vocal tract differences. Our research examines how this perceptual adaptation occurs, using an artificial accent that shifts the typical English weighting of F0 and VOT in signaling stop voicing. In this accent, F0, not VOT, distinguishes voicing. Feedback in the training experiment teaches listeners that VOT is uninformative while F0 is informative. We test the hypothesis that the predictive strength of cue values during training affects learning, such that extreme (a priori more predictive) VOT values promote greater VOT downweighting, while close-together (a priori less predictive) F0 values promote greater F0 upweighting. Participants undergo training with highly or weakly predictive F0 and VOT values, resulting in four groups. Training includes identification of a spoken word given options of either a voiced or voiceless initial consonant (e.g. “bear” or “pear”), followed by feedback meant to teach listeners to pay attention only to F0 and not to VOT. Participants are then tested on a greater variety of F0, VOT, and place of articulation values to measure generalization. We are currently collecting data towards a planned 40 participants per group (160 total).
Clement et al. (Wed,) studied this question.