A unified view for discriminative objective functions based on negative exponential of difference measure between strings

Atsushi Nakamura, Erik McDermott, Shinji Watanabe, Shigeru Katagiri

Research output: Chapter in Book/Report/Conference proceedingConference contribution

11 Citations (Scopus)

Abstract

This paper presents a novel unified view of a wide variety of objective functions suitable for discriminative training applied to sequential pattern recognition problems, such as automatic speech recognition. Focusing on a central component of conventional objective functions, the sum of modified joint probabilities of observations and strings, the analysis generalizes these objective functions by weighting each term in the sum by an important function, the negative exponential of difference measure between strings. The interesting and valuable results of this investigation are highlighted in a comprehensive relationship chart that covers all of the common approaches (Maximum Mutual Information, Minimum Classification Error, Minimum Phone/Word Error), as well as corresponding novel generalizations and modifications of those approaches.

Original languageEnglish
Title of host publication2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009
Pages1633-1636
Number of pages4
DOIs
Publication statusPublished - 2009
Externally publishedYes
Event2009 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2009 - Taipei
Duration: 2009 Apr 192009 Apr 24

Other

Other2009 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2009
CityTaipei
Period09/4/1909/4/24

Fingerprint

Speech recognition
Pattern recognition

Keywords

  • Automatic speech recognition
  • Discriminative training
  • Generalized objective function
  • Laplace-Stieltjes transform
  • Negative exponential function

ASJC Scopus subject areas

  • Signal Processing
  • Software
  • Electrical and Electronic Engineering

Cite this

Nakamura, A., McDermott, E., Watanabe, S., & Katagiri, S. (2009). A unified view for discriminative objective functions based on negative exponential of difference measure between strings. In 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009 (pp. 1633-1636). [4959913] https://doi.org/10.1109/ICASSP.2009.4959913

A unified view for discriminative objective functions based on negative exponential of difference measure between strings. / Nakamura, Atsushi; McDermott, Erik; Watanabe, Shinji; Katagiri, Shigeru.

2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. p. 1633-1636 4959913.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Nakamura, A, McDermott, E, Watanabe, S & Katagiri, S 2009, A unified view for discriminative objective functions based on negative exponential of difference measure between strings. in 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009., 4959913, pp. 1633-1636, 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2009, Taipei, 09/4/19. https://doi.org/10.1109/ICASSP.2009.4959913
Nakamura A, McDermott E, Watanabe S, Katagiri S. A unified view for discriminative objective functions based on negative exponential of difference measure between strings. In 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. p. 1633-1636. 4959913 https://doi.org/10.1109/ICASSP.2009.4959913
Nakamura, Atsushi ; McDermott, Erik ; Watanabe, Shinji ; Katagiri, Shigeru. / A unified view for discriminative objective functions based on negative exponential of difference measure between strings. 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. pp. 1633-1636
@inproceedings{6f462e0669e74978bfa205552d671a5c,
title = "A unified view for discriminative objective functions based on negative exponential of difference measure between strings",
abstract = "This paper presents a novel unified view of a wide variety of objective functions suitable for discriminative training applied to sequential pattern recognition problems, such as automatic speech recognition. Focusing on a central component of conventional objective functions, the sum of modified joint probabilities of observations and strings, the analysis generalizes these objective functions by weighting each term in the sum by an important function, the negative exponential of difference measure between strings. The interesting and valuable results of this investigation are highlighted in a comprehensive relationship chart that covers all of the common approaches (Maximum Mutual Information, Minimum Classification Error, Minimum Phone/Word Error), as well as corresponding novel generalizations and modifications of those approaches.",
keywords = "Automatic speech recognition, Discriminative training, Generalized objective function, Laplace-Stieltjes transform, Negative exponential function",
author = "Atsushi Nakamura and Erik McDermott and Shinji Watanabe and Shigeru Katagiri",
year = "2009",
doi = "10.1109/ICASSP.2009.4959913",
language = "English",
isbn = "9781424423545",
pages = "1633--1636",
booktitle = "2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009",

}

TY - GEN

T1 - A unified view for discriminative objective functions based on negative exponential of difference measure between strings

AU - Nakamura, Atsushi

AU - McDermott, Erik

AU - Watanabe, Shinji

AU - Katagiri, Shigeru

PY - 2009

Y1 - 2009

N2 - This paper presents a novel unified view of a wide variety of objective functions suitable for discriminative training applied to sequential pattern recognition problems, such as automatic speech recognition. Focusing on a central component of conventional objective functions, the sum of modified joint probabilities of observations and strings, the analysis generalizes these objective functions by weighting each term in the sum by an important function, the negative exponential of difference measure between strings. The interesting and valuable results of this investigation are highlighted in a comprehensive relationship chart that covers all of the common approaches (Maximum Mutual Information, Minimum Classification Error, Minimum Phone/Word Error), as well as corresponding novel generalizations and modifications of those approaches.

AB - This paper presents a novel unified view of a wide variety of objective functions suitable for discriminative training applied to sequential pattern recognition problems, such as automatic speech recognition. Focusing on a central component of conventional objective functions, the sum of modified joint probabilities of observations and strings, the analysis generalizes these objective functions by weighting each term in the sum by an important function, the negative exponential of difference measure between strings. The interesting and valuable results of this investigation are highlighted in a comprehensive relationship chart that covers all of the common approaches (Maximum Mutual Information, Minimum Classification Error, Minimum Phone/Word Error), as well as corresponding novel generalizations and modifications of those approaches.

KW - Automatic speech recognition

KW - Discriminative training

KW - Generalized objective function

KW - Laplace-Stieltjes transform

KW - Negative exponential function

UR - http://www.scopus.com/inward/record.url?scp=70349205533&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70349205533&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2009.4959913

DO - 10.1109/ICASSP.2009.4959913

M3 - Conference contribution

SN - 9781424423545

SP - 1633

EP - 1636

BT - 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009

ER -