DISCRIMINATIVE HMMS, LOG-LINEAR MODELS, AND CRFS: WHAT IS THE DIFFERENCE?

被引：7

作者：

Heigold, G. ^{[1
]}

Wiesler, S. ^{[1
]}

Nussbaum-Thom, M. ^{[1
]}

Lehnen, P. ^{[1
]}

Schlueter, R. ^{[1
]}

Ney, H. ^{[1
]}

机构：

[1] Rhein Westfal TH Aachen, Dept Comp Sci, Aachen, Germany

来源：

2010 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING | 2010年

关键词：

speech recognition; hidden Markov model; discriminative training; log-linear model; conditional random field;

D O I：

10.1109/ICASSP.2010.5495228

中图分类号：

O42 [声学];

学科分类号：

070206 ; 082403 ;

摘要：

Recently, there have been many papers studying discriminative acoustic modeling techniques like conditional random fields or discriminative training of conventional Gaussian HMMs. This paper will give an overview of the recent work and progress. We will strictly distinguish between the type of acoustic models on the one hand and the training criterion on the other hand. We will address two issues in more detail: the relation between conventional Gaussian HMMs and conditional random fields and the advantages of formulating the training criterion as a convex optimization problem. Experimental results for various speech tasks will be presented to carefully evaluate the different concepts and approaches, including both a digit string and large vocabulary continuous speech recognition tasks.

引用

页码：5546 / 5549

页数：4

共 14 条

[1]

Abdel-Haleem Y. H., 2006, THESIS U SHEFFIELD F

[2]

Gunawardana A., 2005, INT LISB PORT SEPT

[3]

Heigold G., 2009, ICASSP TAIP TAIW APR

[4]

Heigold G., 2009, INT BRIGHT ENGL SEPT

[5]

Heigold G., 2008, ICML HELS FINL JUL

[6]

Heigold G., 2008, INT BRISB AUSTR SEPT

[7] Maximum entropy direct models for speech recognition [J].

Kuo, HKJ ;

Gao, YQ .

IEEE TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING, 2006, 14 (03) :873-881

[8]

Layton M.I., 2006, ICASSP TOUL FRANC MA

[9]

Morris J., 2009, INT BRIGHT ENGL SEPT

[10]

RIEDMILLER, 1993, P ICNN 93 SAN FRANC

← 1 2 →