Structured phonetic variation facilitates talker identification

Abstract

Individual talkers systematically differ in their phonetic implementation of speech sounds, and listeners use this structure to facilitate language comprehension. Here we test the hypothesis that listeners’ sensitivity to talker-specific phonetic variation also facilitates voice processing. Listeners completed training and test phases. During training, listeners learned to associate talkers’ voices with cartoon avatars (with feedback). Training stimuli consisted of single-word utterances from two minimal pairs (gain, cane, goal, and coal). In one condition, voice-onset-times (VOTs) of the voiceless-initial words were structured such that each talker had a characteristic VOT. In the other condition, listeners heard the same VOTs, but no talker-specific phonetic structure was present. At test, listeners heard all VOT variants for trained and novel words, and they performed the same voice-avatar identification task (without feedback). The results showed that given limited exposure (one training block), learning of the talkers’ voices was equivalent between the two conditions. However, given extended exposure (three training blocks), talker identification was improved with respect to both accuracy and processing time for listeners who received structured phonetic variation compared to those who did not. These findings suggest that given sufficient time to learn talkers’ idiolects, listeners use structured phonetic variation to facilitate voice processing.