Tempo-based segregation of spoken sentences

Abstract

The ability to make use of differences in speech rhythms to selectively attend to a single spoken message in a multi-talker background was examined in a series of studies. Sentences from the coordinate response measure corpus provided a set of stimuli with a common rhythmic framework spoken by several talkers at similar speaking rates. Subjects were asked to identify two key words spoken in a “target” sentence identified by a word (call sign) near the beginning of the sentence. The target talker was always in the same male voice and either two or six background talkers were presented in different voices (half male and half female). The rate of the background talkers was manipulated to create natural sounding speech that preserved the original pitch and speech rhythms at faster and slower speaking rates. Unaltered target sentences were presented in the presence of faster, unaltered, or slower competing sentences. Performance was poorest with matching target and background tempos, with substantial increases in performance as the tempo differences increased. Modification of the target-sentence rate confirmed that the effect is due to the relative timing of target and background speech, rather than the properties of rate-modified background speech. [Work supported by NIH-NIA.]