Predicting Who Will Be the Next Speaker and When in Multi-party Meetings

An understanding of the mechanisms involved in face-to-face communication will contribute to designing advanced video conferencing and dialogue systems. Turn-taking, the situation where the speaker changes, is especially important in multi-party meetings. For smooth turn-taking, the participants need to predict who will start speaking next and to consider a strategy for achieving good timing to speak next. Our aim is to clarify the kinds of behavior that contribute to smooth turn-taking and to develop a model for predicting the next speaker and the start time of the next speaker’s utterance in multi-party meetings. We focus on gaze behavior and respiration near the end of the current speaker’s utterance. We empirically demonstrate that gaze behavior and respiration have a relation to the next speaker and the start timing of the next utterance in multi-party meetings. A prediction model based on the results reveals that gaze behavior and respiration contribute to predicting the next speaker and the timing of the next utterance. © 2015 Nippon Telegraph and Telephone Corp.. All rights reserved.