Empirical error rate minimization based linear discriminant analysis

Hung Shin Lee, Berlin Chen

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

7 Citations (Scopus)

Abstract

Linear discriminant analysis (LDA) is designed to seek a linear transformation that projects a data set into a lower-dimensional feature space while retaining geometrical class separability. However, LDA cannot always guarantee better classification accuracy. One of the possible reasons lies in that its formulation is not directly associated with the classification error rate, so that it is not necessarily suited for the allocation rule governed by a given classifier, such as that employed in automatic speech recognition (ASR). In this paper, we extend the classical LDA by leveraging the relationship between the empirical classification error rate and the Mahalanobis distance for each respective class pair, and modify the original between-class scatter from a measure of the squared Euclidean distance to the pairwise empirical classification accuracy for each class pair, while preserving the lightweight solvability and taking no distributional assumption, just as what LDA does. Experimental results seem to demonstrate that our approach yields moderate improvements over LDA on the large vocabulary continuous speech recognition (LVCSR) task.

Original languageEnglish
Title of host publication2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009
Pages1801-1804
Number of pages4
DOIs
Publication statusPublished - 2009 Sep 23
Event2009 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2009 - Taipei, Taiwan
Duration: 2009 Apr 192009 Apr 24

Publication series

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

Other

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

Fingerprint

Discriminant analysis
Continuous speech recognition
Linear transformations
Speech recognition
Classifiers

Keywords

  • Feature extraction
  • Pattern classification
  • Speech recognition

ASJC Scopus subject areas

  • Software
  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Lee, H. S., & Chen, B. (2009). Empirical error rate minimization based linear discriminant analysis. In 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009 (pp. 1801-1804). [4959955] (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings). https://doi.org/10.1109/ICASSP.2009.4959955

Empirical error rate minimization based linear discriminant analysis. / Lee, Hung Shin; Chen, Berlin.

2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. p. 1801-1804 4959955 (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings).

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

Lee, HS & Chen, B 2009, Empirical error rate minimization based linear discriminant analysis. in 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009., 4959955, ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, pp. 1801-1804, 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2009, Taipei, Taiwan, 09/4/19. https://doi.org/10.1109/ICASSP.2009.4959955
Lee HS, Chen B. Empirical error rate minimization based linear discriminant analysis. In 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. p. 1801-1804. 4959955. (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings). https://doi.org/10.1109/ICASSP.2009.4959955
Lee, Hung Shin ; Chen, Berlin. / Empirical error rate minimization based linear discriminant analysis. 2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009. 2009. pp. 1801-1804 (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings).
@inproceedings{7c29d630d2a84329a695715cca8476f0,
title = "Empirical error rate minimization based linear discriminant analysis",
abstract = "Linear discriminant analysis (LDA) is designed to seek a linear transformation that projects a data set into a lower-dimensional feature space while retaining geometrical class separability. However, LDA cannot always guarantee better classification accuracy. One of the possible reasons lies in that its formulation is not directly associated with the classification error rate, so that it is not necessarily suited for the allocation rule governed by a given classifier, such as that employed in automatic speech recognition (ASR). In this paper, we extend the classical LDA by leveraging the relationship between the empirical classification error rate and the Mahalanobis distance for each respective class pair, and modify the original between-class scatter from a measure of the squared Euclidean distance to the pairwise empirical classification accuracy for each class pair, while preserving the lightweight solvability and taking no distributional assumption, just as what LDA does. Experimental results seem to demonstrate that our approach yields moderate improvements over LDA on the large vocabulary continuous speech recognition (LVCSR) task.",
keywords = "Feature extraction, Pattern classification, Speech recognition",
author = "Lee, {Hung Shin} and Berlin Chen",
year = "2009",
month = "9",
day = "23",
doi = "10.1109/ICASSP.2009.4959955",
language = "English",
isbn = "9781424423545",
series = "ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings",
pages = "1801--1804",
booktitle = "2009 IEEE International Conference on Acoustics, Speech, and Signal Processing - Proceedings, ICASSP 2009",

}

TY - GEN

T1 - Empirical error rate minimization based linear discriminant analysis

AU - Lee, Hung Shin

AU - Chen, Berlin

PY - 2009/9/23

Y1 - 2009/9/23

N2 - Linear discriminant analysis (LDA) is designed to seek a linear transformation that projects a data set into a lower-dimensional feature space while retaining geometrical class separability. However, LDA cannot always guarantee better classification accuracy. One of the possible reasons lies in that its formulation is not directly associated with the classification error rate, so that it is not necessarily suited for the allocation rule governed by a given classifier, such as that employed in automatic speech recognition (ASR). In this paper, we extend the classical LDA by leveraging the relationship between the empirical classification error rate and the Mahalanobis distance for each respective class pair, and modify the original between-class scatter from a measure of the squared Euclidean distance to the pairwise empirical classification accuracy for each class pair, while preserving the lightweight solvability and taking no distributional assumption, just as what LDA does. Experimental results seem to demonstrate that our approach yields moderate improvements over LDA on the large vocabulary continuous speech recognition (LVCSR) task.

AB - Linear discriminant analysis (LDA) is designed to seek a linear transformation that projects a data set into a lower-dimensional feature space while retaining geometrical class separability. However, LDA cannot always guarantee better classification accuracy. One of the possible reasons lies in that its formulation is not directly associated with the classification error rate, so that it is not necessarily suited for the allocation rule governed by a given classifier, such as that employed in automatic speech recognition (ASR). In this paper, we extend the classical LDA by leveraging the relationship between the empirical classification error rate and the Mahalanobis distance for each respective class pair, and modify the original between-class scatter from a measure of the squared Euclidean distance to the pairwise empirical classification accuracy for each class pair, while preserving the lightweight solvability and taking no distributional assumption, just as what LDA does. Experimental results seem to demonstrate that our approach yields moderate improvements over LDA on the large vocabulary continuous speech recognition (LVCSR) task.

KW - Feature extraction

KW - Pattern classification

KW - Speech recognition

UR - http://www.scopus.com/inward/record.url?scp=68149124028&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=68149124028&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2009.4959955

DO - 10.1109/ICASSP.2009.4959955

M3 - Conference contribution

AN - SCOPUS:68149124028

SN - 9781424423545

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

SP - 1801

EP - 1804

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

ER -