10 - Word and Phrase Recognition in Speech Processing

Abstract

This chapter discusses the later stages in speech recognition within the information-processing model. An information-processing model of speech recognition provides one method of studying the cognitive processes that are involved in the analysis of a spoken message. In the information-processing approach to speech recognition, the analysis of a spoken message is viewed from the perspective of information, which is transformed in a sequence of stages, from the initial level of the acoustic signal to the final level of meaning. Each level of information in the sequence of stages corresponds to a particular structure in the information-processing model. Primary consideration is given to the process of word and phrase recognition, which is characterized in the speech model as the initial identification of words and familiar phrases from information in synthesized auditory memory. The chapter discusses a model of word and phrase recognition in which the listener attempts to identify the words of a message individually as they are presented. It is suggested that in normal speech, previous context usually allows such a word-by-word analysis. However, when immediate recognition is not possible, such as for the first words of a sentence, the listener delays the identification process by several words within the constraints of auditory memory. Several factors are noted that could account for the listener's systematic errors. These included non-perceptual factors, such as response bias and acoustic cues to grammatical structure, and the grammatical structure of the sentence. Systematic errors in click localization are influenced by analyses of message meaning occurring after word identification.