The effect of filtered speech on speaker race and sex identifications

Abstract

The purpose of this investigation was to determine the relative importance of portions of the broadband speech spectrum in speaker height and weight identification in an attempt to provide more evidence on the specific acoustic cues that function in such identification tasks. A total of 30 speakers, 15 females and 15 males, recorded a standard prose passage. Three master tapes were constructed from these recordings, representing the three experimental conditions in the study: unfiltered, 255 Hz low-pass filtered, and 255 Hz high-pass filtered tapes. A total of 30 judges participated in three sessions, one for each of the three master tapes. In each session they were asked to make direct estimations of the height and weight of each speaker as well as to indicate the overall confidence for their height and weight judgments at the end of each session by means of a seven-point confidence rating scale. The order of presentation of the tapes was counterbalanced so that six orders were used, with five judges assigned to each of the six order groups, Results indicate that speaker height and weight identification accuracy is not significantly affected by filtering of the speech signal. Implications of these findings and suggestions for future research are discussed.