Maximum model distance discriminative training for text-independent speaker verification

Q. Y. HONG; S. KWONG

doi:10.1109/IECON.2004.1431850

Maximum model distance discriminative training for text-independent speaker verification

Q. Y. HONG, S. KWONG

Research output: Book Chapters | Papers in Conference Proceedings › Conference paper (refereed) › Research › peer-review

1 Citation (Scopus)

Abstract

This paper presents the design and implementation of text-independent speaker verification. We apply the maximum model distance (MMD) algorithm to the Gaussian mixture model (GMM) training. The traditional maximum likelihood (ML) method only utilizes the labeled utterances for each speaker model, which probably leads to a local optimization solution. By maximizing the model distance between the target and competing speakers, MMD could add the discriminative capability into the training procedure and then improve the verification performance. Based on the TIMIT corpus, we designed the verification experiments and the results show that the equal error rate (EER) could be reduced greatly compared with the traditional ML method. © 2004 IEEE.

Original language	English
Title of host publication	IECON 2004 : 30th Annual Conference of IEEE Industrial Electronics Society
Publisher	IEEE
Pages	1769-1774
Number of pages	6
Volume	2
ISBN (Print)	0780387309
DOIs	https://doi.org/10.1109/IECON.2004.1431850
Publication status	Published - 2004
Externally published	Yes
Event	30th Annual Conference of IEEE Industrial Electronics Society, 2004 - Busan, Korea, Republic of Duration: 2 Nov 2004 → 6 Nov 2004

Conference

Conference	30th Annual Conference of IEEE Industrial Electronics Society, 2004
Abbreviated title	IECON 2004
Country/Territory	Korea, Republic of
City	Busan
Period	2/11/04 → 6/11/04

Access to Document

10.1109/IECON.2004.1431850

Cite this

@inproceedings{8440ac3b58164e4ba48b79afb6b1e911,

title = "Maximum model distance discriminative training for text-independent speaker verification",

abstract = "This paper presents the design and implementation of text-independent speaker verification. We apply the maximum model distance (MMD) algorithm to the Gaussian mixture model (GMM) training. The traditional maximum likelihood (ML) method only utilizes the labeled utterances for each speaker model, which probably leads to a local optimization solution. By maximizing the model distance between the target and competing speakers, MMD could add the discriminative capability into the training procedure and then improve the verification performance. Based on the TIMIT corpus, we designed the verification experiments and the results show that the equal error rate (EER) could be reduced greatly compared with the traditional ML method. {\textcopyright} 2004 IEEE.",

author = "HONG, {Q. Y.} and S. KWONG",

year = "2004",

doi = "10.1109/IECON.2004.1431850",

language = "English",

isbn = "0780387309",

volume = "2",

pages = "1769--1774",

booktitle = "IECON 2004 : 30th Annual Conference of IEEE Industrial Electronics Society",

publisher = "IEEE",

note = "30th Annual Conference of IEEE Industrial Electronics Society, 2004, IECON 2004 ; Conference date: 02-11-2004 Through 06-11-2004",

}

HONG, QY & KWONG, S 2004, Maximum model distance discriminative training for text-independent speaker verification. in IECON 2004 : 30th Annual Conference of IEEE Industrial Electronics Society . vol. 2, IEEE, pp. 1769-1774, 30th Annual Conference of IEEE Industrial Electronics Society, 2004, Busan, Korea, Republic of, 2/11/04. https://doi.org/10.1109/IECON.2004.1431850

TY - GEN

T1 - Maximum model distance discriminative training for text-independent speaker verification

AU - HONG, Q. Y.

AU - KWONG, S.

PY - 2004

Y1 - 2004

N2 - This paper presents the design and implementation of text-independent speaker verification. We apply the maximum model distance (MMD) algorithm to the Gaussian mixture model (GMM) training. The traditional maximum likelihood (ML) method only utilizes the labeled utterances for each speaker model, which probably leads to a local optimization solution. By maximizing the model distance between the target and competing speakers, MMD could add the discriminative capability into the training procedure and then improve the verification performance. Based on the TIMIT corpus, we designed the verification experiments and the results show that the equal error rate (EER) could be reduced greatly compared with the traditional ML method. © 2004 IEEE.

AB - This paper presents the design and implementation of text-independent speaker verification. We apply the maximum model distance (MMD) algorithm to the Gaussian mixture model (GMM) training. The traditional maximum likelihood (ML) method only utilizes the labeled utterances for each speaker model, which probably leads to a local optimization solution. By maximizing the model distance between the target and competing speakers, MMD could add the discriminative capability into the training procedure and then improve the verification performance. Based on the TIMIT corpus, we designed the verification experiments and the results show that the equal error rate (EER) could be reduced greatly compared with the traditional ML method. © 2004 IEEE.

UR - http://www.scopus.com/inward/record.url?scp=20544449723&partnerID=8YFLogxK

U2 - 10.1109/IECON.2004.1431850

DO - 10.1109/IECON.2004.1431850

M3 - Conference paper (refereed)

SN - 0780387309

VL - 2

SP - 1769

EP - 1774

BT - IECON 2004 : 30th Annual Conference of IEEE Industrial Electronics Society

PB - IEEE

T2 - 30th Annual Conference of IEEE Industrial Electronics Society, 2004

Y2 - 2 November 2004 through 6 November 2004

ER -

Maximum model distance discriminative training for text-independent speaker verification

Abstract

Conference

Access to Document

Other files and links

Fingerprint

Cite this