Analysis of Emphasis and Prosodic Features on Face-to-Face Discussions

We focus on recognizing speakers' emphasized sections during face-to-face discussions. There is a problem that a speaker’ s emphasized sections are not necessarily conveyed to listeners. Thus, we compared sections that speakers intend as emphasized with sections that listeners perceive as emphasized. We also analyzed the influence of prosodic features on emphasized section recognition. Two types of discussion were designed for our experiment. One is a free discussion and the other is a scripted discussion that contains the direction of emphasized sections. The utterances were recorded during the discussions. The emphasized sections were answered by listener individually after listening to the recorded experiment. After the experiment, the percentage of correspondence of the emphasized sections between the speakers and listeners was calculated. The results showed that there was an insignificant difference in the percentage of correspondences of emphasized sections between the free and scripted discussions. It was difficult for speakers to convey the emphasized sections to the listeners correctly, even if the emphasized sections were directed in advance. The emphasized sections could not be estimated from the pitch and power standard deviations clearly. Finally, we analyzed the reasons listeners recognized emphasized sections, Key words of affirmation.