| Popis: |
This research assessed the nature of acoustic information used in talker normalization and speaker recognition, evaluated three models of talker normalization, and investigated the relationship between the mechanism(s) used by talker normalization and speaker recognition. Two experiments were conducted. Experiment 1 was an AX (same/different) verification task designed to assess the ability of listeners to discriminate eight talkers in three stimulus conditions. The results suggested that listeners are consistent in their confusions across stimulus conditions and can use relatively steady‐state acoustic information to discriminate the talkers. The confusion matrices from experiment 1 were used to create three talker normalization conditions in experiment 2: single talker, confusable talkers, and discriminable talkers. Listeners in experiment 2 were tested in an AX (same/different) phoneme discrimination task. The results suggest that relatively dynamic information is important for talker normalization and that the contextual tuning model is the most appropriate conceptualization of the talker normalization process. The results are discussed with respect to the relationship between talker normalization and speaker verification. |
