Author: Sonya M. Sheffert, David B. Pisoni, Nathan Large, Jennifer Fellowes and Robert Remez
Abstract:
This study explores the learning and generalization of voices using sinewave
speech patterns. The present experiment was designed to address an asymmetry in
generalization performance reported in a previous perceptual learning study
(Sheffert, Pisoni, Fellowes & Remez, 1996). In that study, listeners who became
familiar with a talker¹s voice through listening to sinewave replicas of naturally
produced sentences showed excellent generalization to both novel sinewave and
natural speech sentences, whereas speaker-specific knowledge acquired through
natural speech only generalized to natural speech sentences. The primary
difference between the two experiments was the amount of experience subjects had
with sinewave materials before the generalization tests. The present experiment
tested the hypothesis that the ability to recognize familiar voices from sinewaves
depends on having prior phonetic experience with the unusual acoustic properties
of sinewave signals. The experiment consisted of three phases: Training,
transcription, and generalization. In the training phase, listeners learned to
categorize ten talkers from naturally produced sentences. In the transcription
phase, listeners were familiarized with sinewaves by transcribing sinewave
sentences produced in an unfamiliar voice. In the generalization phase, listeners¹
ability to recognize the ten talkers from a novel naturally produced sentences and
sinewave replicas was assessed. The results confirmed the earlier findings of
Sheffert et al. (1996) by demonstrating that speaker-specific knowledge acquired
during the perceptual training task generalized readily to novel natural
utterance, but not to novel sinewave utterances. The data also show that prior
exposure to the unusual nonspeech tonal patterns did not improve generalization
performance. This pattern of results demonstrates that subjects¹ ability to
exploit the talker-specific phonetic information present in the sinewave replicas
does not depend on having phonetic experience with sinewave speech patterns.