Abstract:

Word perception can be modeled as a two-stage speech recognition process involving, first, recognition of individual phonemes and, second, substitution of missing elements using contextual information. Within this conceptual framework, contextual information may be acoustic in nature, coarticulation, and it may be related to lexical context. Using nonsense CVC syllables, we have analyzed the effects of coarticulation; using meaningful CVC syllables, the additional effects of lexical context. Supplementary estimates of the effects of lexical context were obtained from an incomplete orthographic presentation of the meaningful CVC syllables. The acoustic materials were presented to normal-hearing subjects and to hearing-impaired subjects. The results show large differences in recognition probability of the individual phonemes at low speech-to-noise ratios and marked contextual effects, attributed to coarticulation, for nonsense CVC syllables presented in noise. In the Dutch language, meaningful CVC syllables appear to consist of 2.5 statistically independent elements (rather than 3), which is attributed to the lexical context effects. Near threshold, response behavior of the hearing impaired is heavily based on lexical information, which lowers this number of statistically independent elements to a value close to 1. [This research was performed in collaboration with Dr. A. J. Bosman and Dr. A. W. Bronkhorst.]