Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector

Ikuya Hirano; Kong Aik Lee; Zhaofeng Zhang; Longbiao Wang; Atsuhiko Kai

doi:10.1109/ISCSLP.2014.6936581

Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector

Ikuya Hirano, Kong Aik Lee, Zhaofeng Zhang, Longbiao Wang, Atsuhiko Kai

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

1 Citation (Scopus)

Abstract

Probabilistic linear discriminant analysis (PLDA) has shown to be an effective model for disentangling speaker and channel variability in the i-vector space for text-independent speaker verification. The speaker and channel subspaces in the PLDA model are typically trained by optimizing the maximum likelihood (ML) criterion. PLDA assumes that i-vectors are normally distributed, which has shown to be violated in practice. This paper advocates the use of discriminative training, in which both target and non-target classes are taken into account to re-train the parameters. The efficacy of the proposed method is confirmed via experiments conducted on common condition 1 and 5 of the core task as specified in the Speaker Recognition Evaluations (SREs) 2010 conducted by the National Institute for Standards and Technology (NIST).

Original language	English
Title of host publication	Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014
Editors	Minghui Dong, Jianhua Tao, Haizhou Li, Thomas Fang Zheng, Yanfeng Lu
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	59-63
Number of pages	5
ISBN (Electronic)	9781479942206
DOIs	https://doi.org/10.1109/ISCSLP.2014.6936581
Publication status	Published - 24 Oct 2014
Externally published	Yes
Event	9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014 - Singapore, Singapore Duration: 12 Sept 2014 → 14 Sept 2014

Publication series

Name	Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014

Conference

Conference	9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014
Country/Territory	Singapore
City	Singapore
Period	12/09/14 → 14/09/14

Keywords

discriminative training
Probabilistic Linear Discriminant Analysis
speaker verification

ASJC Scopus subject areas

Information Systems
Computer Science Applications
Software

More information

10.1109/ISCSLP.2014.6936581

Cite this

Hirano, I., Lee, K. A., Zhang, Z., Wang, L., & Kai, A. (2014). Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector. In M. Dong, J. Tao, H. Li, T. F. Zheng, & Y. Lu (Eds.), Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014 (pp. 59-63). Article 6936581 (Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISCSLP.2014.6936581

Hirano, Ikuya ; Lee, Kong Aik ; Zhang, Zhaofeng et al. / Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector. Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014. editor / Minghui Dong ; Jianhua Tao ; Haizhou Li ; Thomas Fang Zheng ; Yanfeng Lu. Institute of Electrical and Electronics Engineers Inc., 2014. pp. 59-63 (Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014).

@inproceedings{802278c920a24e31830d669d3004098c,

title = "Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector",

abstract = "Probabilistic linear discriminant analysis (PLDA) has shown to be an effective model for disentangling speaker and channel variability in the i-vector space for text-independent speaker verification. The speaker and channel subspaces in the PLDA model are typically trained by optimizing the maximum likelihood (ML) criterion. PLDA assumes that i-vectors are normally distributed, which has shown to be violated in practice. This paper advocates the use of discriminative training, in which both target and non-target classes are taken into account to re-train the parameters. The efficacy of the proposed method is confirmed via experiments conducted on common condition 1 and 5 of the core task as specified in the Speaker Recognition Evaluations (SREs) 2010 conducted by the National Institute for Standards and Technology (NIST).",

keywords = "discriminative training, Probabilistic Linear Discriminant Analysis, speaker verification",

author = "Ikuya Hirano and Lee, {Kong Aik} and Zhaofeng Zhang and Longbiao Wang and Atsuhiko Kai",

note = "Publisher Copyright: {\textcopyright} 2014 IEEE.; 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014 ; Conference date: 12-09-2014 Through 14-09-2014",

year = "2014",

month = oct,

day = "24",

doi = "10.1109/ISCSLP.2014.6936581",

language = "English",

series = "Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "59--63",

editor = "Minghui Dong and Jianhua Tao and Haizhou Li and Zheng, {Thomas Fang} and Yanfeng Lu",

booktitle = "Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014",

}

Hirano, I, Lee, KA, Zhang, Z, Wang, L & Kai, A 2014, Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector. in M Dong, J Tao, H Li, TF Zheng & Y Lu (eds), Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014., 6936581, Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014, Institute of Electrical and Electronics Engineers Inc., pp. 59-63, 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014, Singapore, Singapore, 12/09/14. https://doi.org/10.1109/ISCSLP.2014.6936581

Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector. / Hirano, Ikuya; Lee, Kong Aik; Zhang, Zhaofeng et al.
Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014. ed. / Minghui Dong; Jianhua Tao; Haizhou Li; Thomas Fang Zheng; Yanfeng Lu. Institute of Electrical and Electronics Engineers Inc., 2014. p. 59-63 6936581 (Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014).

Research output: Chapter in book / Conference proceeding › Conference article published in proceeding or book › Academic research › peer-review

TY - GEN

T1 - Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector

AU - Hirano, Ikuya

AU - Lee, Kong Aik

AU - Zhang, Zhaofeng

AU - Wang, Longbiao

AU - Kai, Atsuhiko

PY - 2014/10/24

Y1 - 2014/10/24

N2 - Probabilistic linear discriminant analysis (PLDA) has shown to be an effective model for disentangling speaker and channel variability in the i-vector space for text-independent speaker verification. The speaker and channel subspaces in the PLDA model are typically trained by optimizing the maximum likelihood (ML) criterion. PLDA assumes that i-vectors are normally distributed, which has shown to be violated in practice. This paper advocates the use of discriminative training, in which both target and non-target classes are taken into account to re-train the parameters. The efficacy of the proposed method is confirmed via experiments conducted on common condition 1 and 5 of the core task as specified in the Speaker Recognition Evaluations (SREs) 2010 conducted by the National Institute for Standards and Technology (NIST).

AB - Probabilistic linear discriminant analysis (PLDA) has shown to be an effective model for disentangling speaker and channel variability in the i-vector space for text-independent speaker verification. The speaker and channel subspaces in the PLDA model are typically trained by optimizing the maximum likelihood (ML) criterion. PLDA assumes that i-vectors are normally distributed, which has shown to be violated in practice. This paper advocates the use of discriminative training, in which both target and non-target classes are taken into account to re-train the parameters. The efficacy of the proposed method is confirmed via experiments conducted on common condition 1 and 5 of the core task as specified in the Speaker Recognition Evaluations (SREs) 2010 conducted by the National Institute for Standards and Technology (NIST).

KW - discriminative training

KW - Probabilistic Linear Discriminant Analysis

KW - speaker verification

UR - http://www.scopus.com/inward/record.url?scp=84912076397&partnerID=8YFLogxK

U2 - 10.1109/ISCSLP.2014.6936581

DO - 10.1109/ISCSLP.2014.6936581

M3 - Conference article published in proceeding or book

AN - SCOPUS:84912076397

T3 - Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014

SP - 59

EP - 63

BT - Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014

A2 - Dong, Minghui

A2 - Tao, Jianhua

A2 - Li, Haizhou

A2 - Zheng, Thomas Fang

A2 - Lu, Yanfeng

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014

Y2 - 12 September 2014 through 14 September 2014

ER -

Hirano I, Lee KA, Zhang Z, Wang L, Kai A. Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector. In Dong M, Tao J, Li H, Zheng TF, Lu Y, editors, Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014. Institute of Electrical and Electronics Engineers Inc. 2014. p. 59-63. 6936581. (Proceedings of the 9th International Symposium on Chinese Spoken Language Processing, ISCSLP 2014). doi: 10.1109/ISCSLP.2014.6936581

Single-sided approach to discriminative PLDA training for text-independent speaker verification without using expanded i-vector

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this