For effective indexing of presentation speech such as lectures and seminars, we explore a novel approach based on detection of the audience's interest level. In this work, we deal with poster presentations and focus on the backchannel responses or reactive tokens, which are frequently observed in poster conversations and presumably used for expressing the audience's interest level. First, we note that the most common reactive token "hai (yes)" is mainly used for acknowledging the speech segments, and that there are specific kinds of reactive tokens which can be used for expressing non-verbal information. Then, we made a prosodic analysis and identified effective combinations of the syllabic and prosodic patterns which express interest and surprise.
Index Terms: prosody, backchannel, reactive token, audio indexing
Cite as: Kawahara, T., Chang, Z.-Q., Takanashi, K. (2010) Analysis on prosodic features of Japanese reactive tokens in poster conversations. Proc. Speech Prosody 2010, paper 057
@inproceedings{kawahara10_speechprosody,
author={Tatsuya Kawahara and Zhi-Qiang Chang and Katsuya Takanashi},
title={{Analysis on prosodic features of Japanese reactive tokens in poster conversations}},
year=2010,
booktitle={Proc. Speech Prosody 2010},
pages={paper 057}
}