ISCA Archive SpeechProsody 2010
ISCA Archive SpeechProsody 2010

Prosody, supporting real-time conversation

Hiroki Oohashi, Tomoko Ohsuga, Yasuo Horiuchi, Hideaki Kikuchi, Akira Ichikawa

We assume that prosody contains information forenoticing segment boundaries, syntactic structures, and turn transitions and enable us to predict these more easily. We examined this assumption using the F0 model. Concretely speaking, as for forenotices, we examined whether or not the F0 model parameters can lead to segment boundaries, dependencies of phrases, and turn transitions. On the other hand, as for predictions, we conducted cognitive experiments on turn-taking by presenting stimulations containing only prosody and not phonological information. As a result, the segment boundaries were exactly forenoticed at an accuracy of about 60%, the dependencies of phrases were done at about 80%, the turn transitions were done at about 70%, and the possibility of predictions about turn transitions was indicated.

Index Terms: real-time conversation, word segmentation, turntaking, syntactic structure, F0 model

Cite as: Oohashi, H., Ohsuga, T., Horiuchi, Y., Kikuchi, H., Ichikawa, A. (2010) Prosody, supporting real-time conversation. Proc. Speech Prosody 2010, paper 095

  author={Hiroki Oohashi and Tomoko Ohsuga and Yasuo Horiuchi and Hideaki Kikuchi and Akira Ichikawa},
  title={{Prosody, supporting real-time conversation}},
  booktitle={Proc. Speech Prosody 2010},
  pages={paper 095}