Measuring speaker normalization of fricatives using three methods

Abstract

Speaker normalization effects using continua from ‘‘sue’’ to ‘‘shoe’’ and three experimental tasks: Identification, goodness ratings, and direct prototype estimation are compared. The continua were formed by concatenating synthetic fricative noises ranging from [s] to [sh] with the vowel portion of ‘‘sue’’ or ‘‘shoe’’ produced by a male and a female speaker. Speaker normalization was measured in the identification task as a boundary shift between responses to the ‘‘male’’ stimuli and the ‘‘female’’ stimuli. Comparison of category boundaries was also used to measure the speaker normalization effect in the goodness rating paradigm, where a stimulus was considered to be in the ‘‘s’’ category when the ‘‘goodness as [s]’’ rating was higher than the ‘‘goodness as [sh]’’ rating. Speaking normalization in the prototype estimation procedure was measured as the shift in the average estimate of the [s] or [sh] prototype. Both measures of category boundary shift showed substantial speaker normalization effects, while the magnitude of the effect in the prototype estimation task, though significant, was three to four times smaller. However, this smaller effect closely mirrors the magnitude of the acoustic difference usually found between male and female fricative. [Work supported by NIH.]