Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Prosodic Characteristics of a Spoken Dialogue for Information Query

Hiroya Fujisaki (1), Sumio Ohno (1), Masafumi Osame (1), Mayumi Sakata (2), Keikichi Hirose (2)

(1) Department of Applied Electronics, Science University of Tokyo Noda, Japan
(2) Department of Electronic Engineering, University of Tokyo, Tokyo, Japan

Prosodic characteristics of a simulated dialogue have been analyzed and compared with those of read speech. The simulated dialogue was produced by referring to a written text, while read speech was produced by reading individual sentences of the same text in isolation. Analysis of fundamental frequencies and speech rates indicated that dialogue-style utterances have larger mean value and standard deviation of fundamental frequencies as well as higher mean value of speech rates than reading-style utterances. Further analysis of the F0 contour parameters extracted by using a quantitative model has revealed that the differences in the F0 characteristics are caused by a higher baseline value of the F0 contours as well as an expanded range of variation in the amplitude of accent commands in the dialogue-style utterances.

Full Paper

Bibliographic reference.  Fujisaki, Hiroya / Ohno, Sumio / Osame, Masafumi / Sakata, Mayumi / Hirose, Keikichi (1994): "Prosodic characteristics of a spoken dialogue for information query", In ICSLP-1994, 1103-1106.