Perceptual learning of synthetic speech

Abstract

Over the course of a lifetime, listeners are likely to encounter many unfamiliar talkers speaking the listeners’ native language. Despite the fact that talkers may differ radically in the way they sound, listeners normally learn to understand the novel speech patterns of unfamiliar talkers without difficulty. Part of learning to understand an unfamiliar talker speaking a familiar language is learning to relate novel acoustic patterns in the unfamiliar speech to existing mental representations of familiar phonetic categories. This process has frequently been described in terms of modifying an existing psychological ‘‘space’’ of mental representations by shifting the focus or weight of attention to linguistically useful aspects of the speech signal. In the experiment reported here, native English listeners were trained to better understand an unfamiliar talker—a computer speech synthesizer. Training to recognize and transcribe English words resulted in a significant increase in listeners’ abilities to understand words and to identify consonants. Training also strongly influenced the distribution of attention to particular features of the speech signal, as indicated by changes in listeners’ judgments of the similarity of consonants after training. Implications for theories of perceptual learning and phonetic categorization will be discussed.