ISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition April 13-16, 2003 Tokyo Institute of Technology, Tokyo, Japan

Newscast Speech Summarization Via Sentence Shortening Based on Prosodic Features

Kiyonori Ohtake (1), Kazuhide Yamamoto (1,2), Yuji Toma (3), Shiro Sado (3), Shigeru Masuyama (3), Seiichi Nakagawa (4)

(1) ATR Spoken Language Translation Research Laboratories, Kyoto, Japan
(2) Department of Electrical Engineering, Nagaoka University of Technology, Niigata, Japan
(3) Department of Knowledge-based Information Engineering, Toyohashi University of Technology, Aichi,(4) Department of Information and Computer Sciences, Toyohashi University of Technology, Japan
(4) Department of Information and Computer Sciences, Toyohashi University of Technology, Japan

This paper presents a speech summarizer that summarizes input speech via several prosodic features, unlike models that use a speech recognizer and conventional summarizing techniques proposed in natural language processing. Our approach analyzes the borders of summary units by employing prosodic features of pitch, power, and pause to summarize the input speech. Our summary generation trial implies robustness against noisy input compared with both a sequential connection model of a speech recognizer and a text summarizer.

Full Paper

Bibliographic reference.  Ohtake, Kiyonori / Yamamoto, Kazuhide / Toma, Yuji / Sado, Shiro / Masuyama, Shigeru / Nakagawa, Seiichi (2003): "Newscast speech summarization via sentence shortening based on prosodic features", in SSPR-2003, paper TAO4.