Invariant acoustic correlates of phonetic features and their potential role in speech perception

Abstract

A version of the theory of acoustic invariance states that are invariant acoustic correlates of each phonetic feature, independent of the context in which the feature appears. Examples from a number of features are given to show that, if the appropriate acoustic attribute is defined for each feature, there is indeed a high degree of acoustic invariance for the feature. The implications of such a theory for models of language acquisition and speech perception are discussed. It is suggested that a realistic model of speech perception must have the capability to identify lexical items both through the extraction of a pattern of invariant acoustic attributes for individual features in the matrix that defines the item and, in situations where these attributes are obscured, through the use of secondary, context-dependent cues. [Supported in part by a grant from NINCDS.]