An application of fuzzy entropy clustering in speaker identification

Dat Tran, Michael Wagner

Research output: A Conference proceeding or a Chapter in BookConference contributionpeer-review

2 Citations (Scopus)

Abstract

This paper proposes an application of fuzzy entropy (FE) clustering to train speaker models for speaker identification. The FE objective function is obtained by adding a FE term to the hard C-means objective function. A multiplication factor of the FE term is introduced as the degree of fuzzy entropy, since it is used to control the fuzzification of the FE term. The FE clustering is used in this paper to cluster speech data of speakers into codebooks, i.e. speaker models for vector quantisation-based text-independent speaker identification systems. Experiments on the TI46 database show that a speaker identification system using the FE algorithm achieves better results than that using the well-known K-means algorithm.

Original languageEnglish
Title of host publicationProceedings of the Fifth Joint Conference on Information Sciences, JCIS 2000, Volume 1
EditorsP.P. Wang
Place of PublicationUnited States
PublisherAssociation for Computing Machinery (ACM)
Pages228-231
Number of pages4
Edition1
ISBN (Electronic)9780964345690
ISBN (Print)0964345692
Publication statusPublished - 2000
EventProceedings of the Fifth Joint Conference on Information Sciences, JCIS 2000 - Atlantic City, NJ, United States
Duration: 27 Feb 20003 Mar 2000

Publication series

NameProceedings of the Joint Conference on Information Sciences
Number1
Volume5

Conference

ConferenceProceedings of the Fifth Joint Conference on Information Sciences, JCIS 2000
CountryUnited States
CityAtlantic City, NJ
Period27/02/003/03/00

Fingerprint Dive into the research topics of 'An application of fuzzy entropy clustering in speaker identification'. Together they form a unique fingerprint.

Cite this