Presentation 2008-12-10
Discriminative Rescoring Based on Minimization of Word Errors for Speech Recognition
Akio KOBAYASHI, Takahiro OKU, Shinichi HOMMA, Shoei SATO, Toru IMAI, Tohru TAKAGI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper describes a novel method of rescoring that reflects tendencies of errors in word hypotheses in speech recognition for transcribing broadcast news. The proposed rescoring assigns penalties to sentence hypotheses according to the recognition error tendencies in the training lattices themselves using a set of weighting factors for feature functions activated by a variety of linguistic contexts. We introduced new techniques to obtain the factors and it is based on the minimization of word errors, which explicitly reduces expected word errors. Moreover, we proposed a new time-dependent-adaptive training scheme, which features similarities among temporal correlated articles of broadcast news. The results of transcribing Japanese broadcast news achieved a word error rate (WER) of 7.4%, which was a 6.3% reduction relative to conventional trigram lattice rescoring.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) lattice rescoring / discriminative training / word error minimization / lanugage model adaptation
Paper # NLC2008-67,SP2008-122
Date of Issue

Conference Information
Committee NLC
Conference Date 2008/12/2(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 Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Discriminative Rescoring Based on Minimization of Word Errors for Speech Recognition
Sub Title (in English)
Keyword(1) lattice rescoring
Keyword(2) discriminative training
Keyword(3) word error minimization
Keyword(4) lanugage model adaptation
1st Author's Name Akio KOBAYASHI
1st Author's Affiliation NHK Science & Technical Research Laboratories()
2nd Author's Name Takahiro OKU
2nd Author's Affiliation NHK Science & Technical Research Laboratories
3rd Author's Name Shinichi HOMMA
3rd Author's Affiliation NHK Science & Technical Research Laboratories
4th Author's Name Shoei SATO
4th Author's Affiliation NHK Science & Technical Research Laboratories
5th Author's Name Toru IMAI
5th Author's Affiliation NHK Science & Technical Research Laboratories
6th Author's Name Tohru TAKAGI
6th Author's Affiliation NHK Science & Technical Research Laboratories
Date 2008-12-10
Paper # NLC2008-67,SP2008-122
Volume (vol) vol.108
Number (no) 337
Page pp.pp.-
#Pages 6
Date of Issue