Article
Conversational Audio-Visual Cues indicating Speaker Switches in Multi-Talker Scenarios (en)
* Presenting author
Abstract:
Conversations among people inevitably involve speaker switches and exchange of visual cues. Visual cues can indicate expected speaker switches before they occur. They can also indicate which partner will speak next, and they include cues for when a conversation can continue. It has been suggested that this information is obtained by predicting when one's turn will end. For example, these cues can be conveyed by facial expressions, head movements, eye gaze and upper body movements of conversation partners intending to interrupt the current talker. Clustering these cues according to their informative value to observers provides crucial information that can be used to animate multi-talker scenarios in virtual environments. Furthermore, we expect that mixing these cues, for instance by manipulating pre-recorded videos, will severely interfere with the ability to predict the next speaker.We will present a comprehensive quantitative analysis of upper body movement behavior, including head and eye movements, indicating speaker switches for a prerecorded triadic conversation, as well as a comparison of the effects of congruent and incongruent cues on speaker switch prediction.