Lexical and nonlexical approaches to intelligibility in speech: An intermediate approach

Abstract

Intelligibility is a continuum with a floor and a ceiling. In this session, we are concerned with different ways of defining and quantifying intelligibility for different acoustic applications. A major break point in the continuum is defined by lexical transcription. For most applications, such as evaluating clinical and non-native speaker populations, the focus is on whether a reasonably adept listener would come up with a correct transcription under ideal conditions. For other applications, as in telecommunications or clear speaking style, we consider the effects of less-than-ideal conditions, cognitive fatigue, or listener pleasure. In general, however, the quantification of intelligibility is either defined by listener estimates, listener transcription or by broad statistical trends. In this paper, we consider an intermediate method of measuring the likely intelligibility of speech using acoustic characteristics specific to syllable shape, but not tied to transcription. Examples from measuring speaker fatigue, clear speaking style, and disordered source characteristics will be provided.