Skip to main content
Open Access Publications from the University of California

Talker-Specific Accent: The Role of Idiolect in the Perception of Accented Speech

  • Author(s): Miller, Rachel Marie
  • Advisor(s): Rosenblum, Lawrence D
  • et al.

Strong evidence suggests that familiarity with talker-specific (idiolectic) information benefits speech perception (e.g., Nygaard & Pisoni, 1998). However, it seems that talker familiarity does not influence the perception of accented speech (Sidaras et al., 2009). This dissertation assesses whether idiolect and accent are both encoded (Goldinger, 1998); or instead, if the perception of accented speech involves a process of normalization (e.g., Halle, 1985). Two sets of experiments examined talker-specific and accent-general influences on the perception of accented speech using a speech alignment methodology. Speech alignment is the tendency of individuals to subtly imitate the speech of a person with whom they are speaking and occurs to native (unaccented) speech (Goldinger, 1998). During Experiment Series 1, native English subjects shadowed a Chinese- or Spanish-accented model producing English words. In Experiment 1a, raters judged whether a model's tokens sounded more similar to the shadowed token or to a different subject's token shadowed after a same-accented model. Results revealed significant talker alignment. In Experiment 1b, raters judged whether shadowed tokens were more similar in accent to models with the same or a different accent, neither of whom was shadowed. Accent alignment results were inconclusive due to a response bias, which seemed to be related to the magnitude of a model's accent as measured in Experiment 1c. Generally, the finding of talker alignment suggests that talker-specific information is encoded during accented speech perception. Experiment Series 2 investigated potential causes for a lack of these findings in the experiments of Sidaras et al. (2009). In Experiment 2a, listeners were trained to shadow or transcribe Spanish-accented models and were tested on either the same or different models. No effects of training were found. In Experiment 2b, listeners were instead trained and tested on native, English talkers. There was a significant effect of training, but no influence of familiar talker and no difference in accuracy between shadowers and transcribers. These overall findings suggest that talker-specific information is encoded during the perception of accented speech, supporting an episodic account of speech perception. However, the nature and interaction of talker-specific and accent-general information remains unresolved.

Main Content
Current View