Effects of Speaking Style on the Perceptual Learning of Novel Voices: A First Report1

This study examined the effects of speaking style on the perceptual learning of novel voices in the laboratory. Listeners participated in a voice learning experiment. In the training phase, listeners were asked to learn the names of either seven male or seven female talkers from samples of citation or hyperarticulated speech. In the test phase, listeners were presented with the same stimuli as in the training phase and were asked to identify the talker, with no feedback. In the sentence generalization phase, listeners were asked to identify the same voices producing new sentences in the same speaking style as that used in the previous phases. In the speaking style generalization phase, listeners were asked to identify the same voices in either the same speaking style as the previous phases or in a novel speaking style. The results showed that female voices were easier to learn in a hyperarticulated speaking style relative to a citation speaking style in the training and test phases. For the male voices, no such effect was observed. In addition, voice identification scores increased from the training to the test phase. However, voice identification scores did not improve in subsequent phases, which lacked the feedback provided during the training phase. In the style generalization phase, training with the female voice hyperarticulated tokens provided a greater advantage in identifying voices in a novel style relative to training with female voice citation tokens. No such effect was observed for listeners trained with the male voices. This gender interaction was further explored in a similarity scaling experiment, using stimuli from the first experiment. Listeners were presented with pairs of stimuli that differed in talker but matched in speaking style and gender. The citation sentences of both male and female talkers were rated as significantly more similar than the hyperarticulated sentences. However, for a subset of the stimuli, the difference in mean similarity for the female citation and hyperarticulated sentences was significantly greater than the corresponding difference in the male voices, indicating that the female talkers may have produced a more perceptually distinct hyperarticulated style than the male talkers. These differences may have contributed to the gender effect observed in learning in the first experiment. Taken together, the results of both experiments show that speaking style exerts a strong influence on the learning of novel voices, but its exact role is unclear given the interaction of speaking style and gender of the talker.