Speaking rate normalization in multiple talker conversations
Funding, application period
Humboldt, 2012 - 2014
Speech is inherently variable. But listeners evidently have little difficulty in hearing the same words when they hear them from different speakers or spoken at a fast or slow rate. A central question in psycholinguistics and phonetics is: how do listeners deal with this kind of variation so effortlessly? The solution to this puzzle is important, not only for understanding how speech processing functions in humans but also for improving spoken language human-machine communication systems. The project will shed light on a number of controversies about how humans recognize words. Specifically it will address the disputed role of episodic memory in speech processing: Do listeners remember detailed speaker-specific information such as speaking rate in order to 'tune in' to a speaker's speech or is speaking rate 'normalized' before words are recognized in a speaker-independent fashion? The proposed project will disentangle these opposing accounts by investigating word recognition in the naturalistic situation of multiple-speaker conversations. In order to do so, perception experiments will be carried out testing the identification of fast and slow speakers' voices and utterances after various types of speaker familiarization. Experiments will be implemented by the use of state-of-the art techniques for analyzing and manipulating timing relations in speech. The effect of listeners' familiarity with fast and slow speakers will then be extended to the perception of these speakers' utterances in dialogue where fast and slow speakers become more alike. The project will thus integrate techniques from psycholinguistics and experimental phonetics to address the question of how listeners' process information from multiple speakers. By so doing, the proposed project seeks an answer to a fundamental question in speech science: how important is it to hear WHO is talking?