The nature of the link between speech perception and production is extensively debated. One question in this debate concerns the possible use of shared representations. Talker representations encode information about particular talkers' speech and are formed through perception. We present three experiments that investigate the potential role of talker representations in speech production as measured by their impact on phonetic convergence to voice fundamental frequency (F0). In Experiment 1, female native Dutch speakers (N=32) performed a baseline reading task, followed by a synchronous speech task where they were instructed to synchronize (temporally) to a pre-recorded model talker. The model talker's voice was pitch-shifted to have high or low F0, and half of the sample performed the task in each F0 condition. Participants' F0 values converged towards that of the model talker. In Experiment 2, we added an exposure task before synchronous speech, during which participants (N=32) were familiarized with the model talker at high or low F0. At test, they heard the model talker at the familiar F0. Finally, in Experiment 3, we reversed the F0 manipulation between exposure and test. Thus, participants (N=32) were familiarized with either high or low F0, but heard the opposite F0 condition during synchronous speech. The data revealed that while congruent and conflicting talker information both diminish convergence effect size, only conflicting talker information reduces the probability of convergence. These results indicate that talker information acquired exclusively through perception is used during production, suggesting shared representations across perception and production.
Ulusahin et al. (Fri,) studied this question.