Presentation 2006/12/14
Minimum Bayes-Risk Decoding using Wordgraph and Key-sentence Extraction
Hiroaki NANJO, Tatsuya KAWAHARA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Speech recognition strategy for open-domain speech understanding and its application to a key-sentence extraction, which is a first step of speech understanding, are presented. Firstly, a weighted word error rate (WWER), which is an evaluation measure giving a weight on errors from a viewpoint of speech understanding, is described. Then, we show a decoding method to minimize WWER based on Minimum Bayes-Risk (MBR) framework. We have investigated N-best list based MBR decoding. In this paper, several wordgraph based MBR decoding algorithms are presented, and then, we show the effectiveness of confusion network based decoding strategy. Finally, we demonstrate that the decoding method works reasonably for improving both WWER and key sentence indexing.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) speech recognition / speech understanding / Minimum Bayes-Risk decoding key sentence extraction / wordgraph
Paper # NLC2006-50,SP2006-106
Date of Issue

Conference Information
Committee SP
Conference Date 2006/12/14(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Speech (SP)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Minimum Bayes-Risk Decoding using Wordgraph and Key-sentence Extraction
Sub Title (in English)
Keyword(1) speech recognition
Keyword(2) speech understanding
Keyword(3) Minimum Bayes-Risk decoding key sentence extraction
Keyword(4) wordgraph
1st Author's Name Hiroaki NANJO
1st Author's Affiliation Faculty of Science and Technology, Ryukoku University()
2nd Author's Name Tatsuya KAWAHARA
2nd Author's Affiliation Graduate School of Informatics, Kyoto University
Date 2006/12/14
Paper # NLC2006-50,SP2006-106
Volume (vol) vol.106
Number (no) 443
Page pp.pp.-
#Pages 6
Date of Issue