Towards optimal bayes decision for speech recognition

Jen Tzung Chien, Chih-Hsien Huang, Koichi Shinoda, Sadaoki Furui

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

12 Citations (Scopus)

Abstract

This paper presents a new speech recognition framework towards fulfilling optimal Bayes decision theory, which is essential for general pattern recognition. The recognition procedure is developed through minimizing the Bayes risk, or equivalently the expected loss due to classification action. Typically, loss function measures the penalty/evidence of choosing a candidate hypothesis. This function was manually specified or empirically calculated. Here, we exploit a novel Bayes loss function via testing the hypotheses whether the classification action produces loss or not. A Bayes factor is derived to measure loss in a statistical and meaningful way. Attractively, Bayes loss function using predictive distributions is robust to the uncertainty of environments. Also, optimizing this Bayes criterion equals to minimizing classification errors of test data. The relation between the minimum classification error (MCE) classifier and the proposed optimal Bayes classifier (OBC) is bridged. Specifically, the logarithm of Bayes factor in OBC is analogous to the misclassification measure in MCE when using predictive distribution as the discriminant function. We accordingly build a robust and discriminative classification for large vocabulary continuous speech recognition. In the experiments on broadcast news transcription, the new OBC rule significantly outperforms traditional maximum a posteriori classification.

Original languageEnglish
Title of host publication2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings
Publication statusPublished - 2006 Dec 1
Event2006 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2006 - Toulouse, France
Duration: 2006 May 142006 May 19

Publication series

NameICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
Volume1
ISSN (Print)1520-6149

Other

Other2006 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2006
CountryFrance
CityToulouse
Period06-05-1406-05-19

Fingerprint

Speech recognition
Classifiers
Continuous speech recognition
Decision theory
Transcription
Pattern recognition
Testing
Experiments

All Science Journal Classification (ASJC) codes

  • Software
  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Chien, J. T., Huang, C-H., Shinoda, K., & Furui, S. (2006). Towards optimal bayes decision for speech recognition. In 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings [1659953] (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings; Vol. 1).
Chien, Jen Tzung ; Huang, Chih-Hsien ; Shinoda, Koichi ; Furui, Sadaoki. / Towards optimal bayes decision for speech recognition. 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings. 2006. (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings).
@inproceedings{54c9d31dd8f34e7aad1b98ea8ead2f2b,
title = "Towards optimal bayes decision for speech recognition",
abstract = "This paper presents a new speech recognition framework towards fulfilling optimal Bayes decision theory, which is essential for general pattern recognition. The recognition procedure is developed through minimizing the Bayes risk, or equivalently the expected loss due to classification action. Typically, loss function measures the penalty/evidence of choosing a candidate hypothesis. This function was manually specified or empirically calculated. Here, we exploit a novel Bayes loss function via testing the hypotheses whether the classification action produces loss or not. A Bayes factor is derived to measure loss in a statistical and meaningful way. Attractively, Bayes loss function using predictive distributions is robust to the uncertainty of environments. Also, optimizing this Bayes criterion equals to minimizing classification errors of test data. The relation between the minimum classification error (MCE) classifier and the proposed optimal Bayes classifier (OBC) is bridged. Specifically, the logarithm of Bayes factor in OBC is analogous to the misclassification measure in MCE when using predictive distribution as the discriminant function. We accordingly build a robust and discriminative classification for large vocabulary continuous speech recognition. In the experiments on broadcast news transcription, the new OBC rule significantly outperforms traditional maximum a posteriori classification.",
author = "Chien, {Jen Tzung} and Chih-Hsien Huang and Koichi Shinoda and Sadaoki Furui",
year = "2006",
month = "12",
day = "1",
language = "English",
isbn = "142440469X",
series = "ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings",
booktitle = "2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings",

}

Chien, JT, Huang, C-H, Shinoda, K & Furui, S 2006, Towards optimal bayes decision for speech recognition. in 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings., 1659953, ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, vol. 1, 2006 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2006, Toulouse, France, 06-05-14.

Towards optimal bayes decision for speech recognition. / Chien, Jen Tzung; Huang, Chih-Hsien; Shinoda, Koichi; Furui, Sadaoki.

2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings. 2006. 1659953 (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings; Vol. 1).

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

TY - GEN

T1 - Towards optimal bayes decision for speech recognition

AU - Chien, Jen Tzung

AU - Huang, Chih-Hsien

AU - Shinoda, Koichi

AU - Furui, Sadaoki

PY - 2006/12/1

Y1 - 2006/12/1

N2 - This paper presents a new speech recognition framework towards fulfilling optimal Bayes decision theory, which is essential for general pattern recognition. The recognition procedure is developed through minimizing the Bayes risk, or equivalently the expected loss due to classification action. Typically, loss function measures the penalty/evidence of choosing a candidate hypothesis. This function was manually specified or empirically calculated. Here, we exploit a novel Bayes loss function via testing the hypotheses whether the classification action produces loss or not. A Bayes factor is derived to measure loss in a statistical and meaningful way. Attractively, Bayes loss function using predictive distributions is robust to the uncertainty of environments. Also, optimizing this Bayes criterion equals to minimizing classification errors of test data. The relation between the minimum classification error (MCE) classifier and the proposed optimal Bayes classifier (OBC) is bridged. Specifically, the logarithm of Bayes factor in OBC is analogous to the misclassification measure in MCE when using predictive distribution as the discriminant function. We accordingly build a robust and discriminative classification for large vocabulary continuous speech recognition. In the experiments on broadcast news transcription, the new OBC rule significantly outperforms traditional maximum a posteriori classification.

AB - This paper presents a new speech recognition framework towards fulfilling optimal Bayes decision theory, which is essential for general pattern recognition. The recognition procedure is developed through minimizing the Bayes risk, or equivalently the expected loss due to classification action. Typically, loss function measures the penalty/evidence of choosing a candidate hypothesis. This function was manually specified or empirically calculated. Here, we exploit a novel Bayes loss function via testing the hypotheses whether the classification action produces loss or not. A Bayes factor is derived to measure loss in a statistical and meaningful way. Attractively, Bayes loss function using predictive distributions is robust to the uncertainty of environments. Also, optimizing this Bayes criterion equals to minimizing classification errors of test data. The relation between the minimum classification error (MCE) classifier and the proposed optimal Bayes classifier (OBC) is bridged. Specifically, the logarithm of Bayes factor in OBC is analogous to the misclassification measure in MCE when using predictive distribution as the discriminant function. We accordingly build a robust and discriminative classification for large vocabulary continuous speech recognition. In the experiments on broadcast news transcription, the new OBC rule significantly outperforms traditional maximum a posteriori classification.

UR - http://www.scopus.com/inward/record.url?scp=33947628249&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33947628249&partnerID=8YFLogxK

M3 - Conference contribution

SN - 142440469X

SN - 9781424404698

T3 - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

BT - 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings

ER -

Chien JT, Huang C-H, Shinoda K, Furui S. Towards optimal bayes decision for speech recognition. In 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings. 2006. 1659953. (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings).