The role of sentential prosody in learning voices

The present experiment was designed to investigate the contribution of sentential prosody to the perceptual learning of talker’s voice. Listeners were trained over a three‐day period to learn talker’s voices from three types of utterances: (1) sentence‐length utterances produced with natural sentential prosody, (2) utterances in which individual words were produced in list format and digitally assembled into complete sentences, and (3) words produced in list format and scrambled to remove semantic and syntactic coherence. Equal numbers of male and female listeners learned a set of ten talker’s voices (five male and five female) and learning was evaluated at each training session. On the fourth day of testing, listeners were given a generalization test that consisted of novel sentence‐length utterances produced with natural sentential prosody. Results indicated that learning curves were comparable across conditions, but the ability to generalize to new utterances depended on the utterance type used during training. Individual differences among both listeners and talkers were also observed. Male listeners identified male voices significantly better than female voices. Female listeners identified male and female voices equally well. These results suggest that the perception of talker identity is dependent both on listener‐ and task‐specific characteristics.