Speeded classification of natural and synthetic speech in a lexical decision task

How fast can human observers recognize isolated words? How is the process of word recognition affected by the quality of the initial acoustic-phonetic input in the speech signal? In this paper we report the results of a lexical decision experiment which examined the response times for classification of natural and synthetic word and nonword stimuli. The results showed an overall increase in response time of 145 ms for synthetic stimuli compared to natural speech. In addition, words were recognized 140 ms faster than matched nonword controls. However, there was no interaction between signal type (natural versus synthetic) and classification response (word versus nonword). These results suggest that differences in perception between natural and synthetic speech lie at early stages of perceptual analysis in which the initial phonetic or segmental representation of the input signal is developed rather than at later stages of lexical access and search where these representations are examined or compared prior to execution of the observer's classification response. The findings will be discussed in terms of the extensive use of synthetic speech stimuli in perceptual experiments and its application in voice response systems used in applied settings. [Supported by NINCDS grant NS-12179.]