Discriminative training of acoustic models for system combination

Yuuki Tachioka, Shinji Watanabe

研究成果: Article

4 引用 (Scopus)

抄録

In discriminative training methods, the objective function is designed to improve the performance of automatic speech recognition with reference to correct labels using a single system. On the other hand, system combination methods, which output refined hypotheses by a majority voting scheme, need to build multiple systems that generate complementary hypotheses. This paper aims to unify the both requirements within a discriminative training framework based on the mutual information criterion. That is, we construct complementary models by optimizing the proposed objective function, which yields to minimize the mutual information with base systems' hypotheses, while maximize that with correct labels, at the same time. We also analyze that this scheme corresponds to weight the training data of a complementary system by considering correct and error tendencies in the base systems, which has close relationship with boosting methods. In addition, the proposed method can practically construct complementary systems by simply extending a lattice-based parameter update algorithm in discriminative training, and can adjust the degree of how much the complementary system outputs are different from base system ones. The experiments on highly noisy speech recognition ('The 2nd CHiME challenge') show the effectiveness of the proposed method, compared with a conventional system combination approach.

元の言語English
ページ(範囲)2355-2359
ページ数5
ジャーナルUnknown Journal
出版物ステータスPublished - 2013
外部発表Yes

Fingerprint

Discriminative Training
Acoustic Model
Speech recognition
Labels
education
Acoustics
acoustics
speech recognition
voting
output
tendencies
Mutual Information
Experiments
requirements
Objective function
Majority Voting
Automatic Speech Recognition
Information Criterion
Output
Boosting

ASJC Scopus subject areas

  • Language and Linguistics
  • Human-Computer Interaction
  • Signal Processing
  • Software
  • Modelling and Simulation

これを引用

Discriminative training of acoustic models for system combination. / Tachioka, Yuuki; Watanabe, Shinji.

:: Unknown Journal, 2013, p. 2355-2359.

研究成果: Article

@article{965cc6bf6066439cb1a25f13d8466057,
title = "Discriminative training of acoustic models for system combination",
abstract = "In discriminative training methods, the objective function is designed to improve the performance of automatic speech recognition with reference to correct labels using a single system. On the other hand, system combination methods, which output refined hypotheses by a majority voting scheme, need to build multiple systems that generate complementary hypotheses. This paper aims to unify the both requirements within a discriminative training framework based on the mutual information criterion. That is, we construct complementary models by optimizing the proposed objective function, which yields to minimize the mutual information with base systems' hypotheses, while maximize that with correct labels, at the same time. We also analyze that this scheme corresponds to weight the training data of a complementary system by considering correct and error tendencies in the base systems, which has close relationship with boosting methods. In addition, the proposed method can practically construct complementary systems by simply extending a lattice-based parameter update algorithm in discriminative training, and can adjust the degree of how much the complementary system outputs are different from base system ones. The experiments on highly noisy speech recognition ('The 2nd CHiME challenge') show the effectiveness of the proposed method, compared with a conventional system combination approach.",
keywords = "Boosting, Discriminative training, Margin training, MMI, System combination",
author = "Yuuki Tachioka and Shinji Watanabe",
year = "2013",
language = "English",
pages = "2355--2359",
journal = "Nuclear Physics A",
issn = "0375-9474",
publisher = "Elsevier",

}

TY - JOUR

T1 - Discriminative training of acoustic models for system combination

AU - Tachioka, Yuuki

AU - Watanabe, Shinji

PY - 2013

Y1 - 2013

N2 - In discriminative training methods, the objective function is designed to improve the performance of automatic speech recognition with reference to correct labels using a single system. On the other hand, system combination methods, which output refined hypotheses by a majority voting scheme, need to build multiple systems that generate complementary hypotheses. This paper aims to unify the both requirements within a discriminative training framework based on the mutual information criterion. That is, we construct complementary models by optimizing the proposed objective function, which yields to minimize the mutual information with base systems' hypotheses, while maximize that with correct labels, at the same time. We also analyze that this scheme corresponds to weight the training data of a complementary system by considering correct and error tendencies in the base systems, which has close relationship with boosting methods. In addition, the proposed method can practically construct complementary systems by simply extending a lattice-based parameter update algorithm in discriminative training, and can adjust the degree of how much the complementary system outputs are different from base system ones. The experiments on highly noisy speech recognition ('The 2nd CHiME challenge') show the effectiveness of the proposed method, compared with a conventional system combination approach.

AB - In discriminative training methods, the objective function is designed to improve the performance of automatic speech recognition with reference to correct labels using a single system. On the other hand, system combination methods, which output refined hypotheses by a majority voting scheme, need to build multiple systems that generate complementary hypotheses. This paper aims to unify the both requirements within a discriminative training framework based on the mutual information criterion. That is, we construct complementary models by optimizing the proposed objective function, which yields to minimize the mutual information with base systems' hypotheses, while maximize that with correct labels, at the same time. We also analyze that this scheme corresponds to weight the training data of a complementary system by considering correct and error tendencies in the base systems, which has close relationship with boosting methods. In addition, the proposed method can practically construct complementary systems by simply extending a lattice-based parameter update algorithm in discriminative training, and can adjust the degree of how much the complementary system outputs are different from base system ones. The experiments on highly noisy speech recognition ('The 2nd CHiME challenge') show the effectiveness of the proposed method, compared with a conventional system combination approach.

KW - Boosting

KW - Discriminative training

KW - Margin training

KW - MMI

KW - System combination

UR - http://www.scopus.com/inward/record.url?scp=84893695671&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84893695671&partnerID=8YFLogxK

M3 - Article

AN - SCOPUS:84893695671

SP - 2355

EP - 2359

JO - Nuclear Physics A

JF - Nuclear Physics A

SN - 0375-9474

ER -