An information-theoretic perspective on feature selection in speaker recognition

Thomas Eriksson, Samuel Kim, Hong Goo Kang, Chungyong Lee

Research output: Contribution to journalArticle

39 Citations (Scopus)

Abstract

This letter studies feature selection in speaker recognition from an information-theoretic view. We closely tie the performance, in terms of the expected classification error probability, to the mutual information between speaker identity and features. Information theory can then help us to make qualitative statements about feature selection and performance. We study various common features used for speaker recognition, such as mel-warped cepstrum coefficients and various parameterizations of linear prediction coefficients. The theory and experiments give valuable insights in feature selection and performance of speaker-recognition applications.

Original languageEnglish
Pages (from-to)500-503
Number of pages4
JournalIEEE Signal Processing Letters
Volume12
Issue number7
DOIs
Publication statusPublished - 2005 Jul 1

Fingerprint

Speaker Recognition
Feature Selection
Feature extraction
Cepstrum
Linear Prediction
Information theory
Error Probability
Coefficient
Tie
Information Theory
Parameterization
Mutual Information
Experiment
Experiments

All Science Journal Classification (ASJC) codes

  • Signal Processing
  • Electrical and Electronic Engineering
  • Applied Mathematics

Cite this

@article{dffb0519f0e44a37b7cc69bf3bdda790,
title = "An information-theoretic perspective on feature selection in speaker recognition",
abstract = "This letter studies feature selection in speaker recognition from an information-theoretic view. We closely tie the performance, in terms of the expected classification error probability, to the mutual information between speaker identity and features. Information theory can then help us to make qualitative statements about feature selection and performance. We study various common features used for speaker recognition, such as mel-warped cepstrum coefficients and various parameterizations of linear prediction coefficients. The theory and experiments give valuable insights in feature selection and performance of speaker-recognition applications.",
author = "Thomas Eriksson and Samuel Kim and Kang, {Hong Goo} and Chungyong Lee",
year = "2005",
month = "7",
day = "1",
doi = "10.1109/LSP.2005.849495",
language = "English",
volume = "12",
pages = "500--503",
journal = "IEEE Signal Processing Letters",
issn = "1070-9908",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
number = "7",

}

An information-theoretic perspective on feature selection in speaker recognition. / Eriksson, Thomas; Kim, Samuel; Kang, Hong Goo; Lee, Chungyong.

In: IEEE Signal Processing Letters, Vol. 12, No. 7, 01.07.2005, p. 500-503.

Research output: Contribution to journalArticle

TY - JOUR

T1 - An information-theoretic perspective on feature selection in speaker recognition

AU - Eriksson, Thomas

AU - Kim, Samuel

AU - Kang, Hong Goo

AU - Lee, Chungyong

PY - 2005/7/1

Y1 - 2005/7/1

N2 - This letter studies feature selection in speaker recognition from an information-theoretic view. We closely tie the performance, in terms of the expected classification error probability, to the mutual information between speaker identity and features. Information theory can then help us to make qualitative statements about feature selection and performance. We study various common features used for speaker recognition, such as mel-warped cepstrum coefficients and various parameterizations of linear prediction coefficients. The theory and experiments give valuable insights in feature selection and performance of speaker-recognition applications.

AB - This letter studies feature selection in speaker recognition from an information-theoretic view. We closely tie the performance, in terms of the expected classification error probability, to the mutual information between speaker identity and features. Information theory can then help us to make qualitative statements about feature selection and performance. We study various common features used for speaker recognition, such as mel-warped cepstrum coefficients and various parameterizations of linear prediction coefficients. The theory and experiments give valuable insights in feature selection and performance of speaker-recognition applications.

UR - http://www.scopus.com/inward/record.url?scp=23744450898&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=23744450898&partnerID=8YFLogxK

U2 - 10.1109/LSP.2005.849495

DO - 10.1109/LSP.2005.849495

M3 - Article

AN - SCOPUS:23744450898

VL - 12

SP - 500

EP - 503

JO - IEEE Signal Processing Letters

JF - IEEE Signal Processing Letters

SN - 1070-9908

IS - 7

ER -