Neural networks based channel compensation for i-vector speaker verification

Wei Rao; Xiong Xiao; Chenglin Xu; Haihua Xu; Kongaik Lee; Eng Siong Chng; Haizhou Li

doi:10.1109/ISCSLP.2016.7918436

Neural networks based channel compensation for i-vector speaker verification

Wei Rao, Xiong Xiao, Chenglin Xu, Haihua Xu, Kongaik Lee, Eng Siong Chng, Haizhou Li

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

1 Citation (Scopus)

Abstract

Linear discriminant analysis (LDA) and Gaussian probabilistic LDA (PLDA) have been shown to effectively suppress channel-and session-variability of i-vectors. But they suffer the following limitations: 1) In LDA, a single linear transformation may not be adequate to describe the nonlinear relationship of features and 2) Gaussian-PLDA assumes the speaker and channel factors follow a Gaussian distribution, but they are actually non-Gaussians. We consider neural networks (NN) as a way to overcome the limitations, that captures the nonlinear relationship of features and does not require prior assumptions. This paper investigates three NN based channel compensation methods: deep metric learning, NN classifier, and deep denoising autoencoder and compares their performance with LDA and PLDA. Experiments conducted on NIST 2010 speaker recognition evaluation suggest that NN-based channel compensation methods are superior to LDA and that the performance of NN classifier is better than that of PLDA under most of common evaluation conditions. Additionally, this paper also helps us understand the relationships among LDA, PLDA, and NN based methods.

Original language	English
Title of host publication	Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016
Editors	Hsin-Min Wang, Qingzhi Hou, Yuan Wei, Tan Lee, Jianguo Wei, Lei Xie, Hui Feng, Jianwu Dang, Jianwu Dang
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781509042937
DOIs	https://doi.org/10.1109/ISCSLP.2016.7918436
Publication status	Published - 2 May 2017
Externally published	Yes
Event	10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016 - Tianjin, China Duration: 17 Oct 2016 → 20 Oct 2016

Publication series

Name	Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016

Conference

Conference	10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016
Country/Territory	China
City	Tianjin
Period	17/10/16 → 20/10/16

Keywords

I-vector
LDA
Neural networks
PLDA
Speaker recognition

ASJC Scopus subject areas

Signal Processing
Computer Vision and Pattern Recognition
Linguistics and Language

More information

10.1109/ISCSLP.2016.7918436

Cite this

Rao, W., Xiao, X., Xu, C., Xu, H., Lee, K., Chng, E. S., & Li, H. (2017). Neural networks based channel compensation for i-vector speaker verification. In H.-M. Wang, Q. Hou, Y. Wei, T. Lee, J. Wei, L. Xie, H. Feng, J. Dang, & J. Dang (Eds.), Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016 Article 7918436 (Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISCSLP.2016.7918436

Rao, Wei ; Xiao, Xiong ; Xu, Chenglin et al. / Neural networks based channel compensation for i-vector speaker verification. Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016. editor / Hsin-Min Wang ; Qingzhi Hou ; Yuan Wei ; Tan Lee ; Jianguo Wei ; Lei Xie ; Hui Feng ; Jianwu Dang ; Jianwu Dang. Institute of Electrical and Electronics Engineers Inc., 2017. (Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016).

@inproceedings{0c0e668bdc4c43ae8d57beb0b4741ecb,

title = "Neural networks based channel compensation for i-vector speaker verification",

abstract = "Linear discriminant analysis (LDA) and Gaussian probabilistic LDA (PLDA) have been shown to effectively suppress channel-and session-variability of i-vectors. But they suffer the following limitations: 1) In LDA, a single linear transformation may not be adequate to describe the nonlinear relationship of features and 2) Gaussian-PLDA assumes the speaker and channel factors follow a Gaussian distribution, but they are actually non-Gaussians. We consider neural networks (NN) as a way to overcome the limitations, that captures the nonlinear relationship of features and does not require prior assumptions. This paper investigates three NN based channel compensation methods: deep metric learning, NN classifier, and deep denoising autoencoder and compares their performance with LDA and PLDA. Experiments conducted on NIST 2010 speaker recognition evaluation suggest that NN-based channel compensation methods are superior to LDA and that the performance of NN classifier is better than that of PLDA under most of common evaluation conditions. Additionally, this paper also helps us understand the relationships among LDA, PLDA, and NN based methods.",

keywords = "I-vector, LDA, Neural networks, PLDA, Speaker recognition",

author = "Wei Rao and Xiong Xiao and Chenglin Xu and Haihua Xu and Kongaik Lee and Chng, {Eng Siong} and Haizhou Li",

note = "Publisher Copyright: {\textcopyright} 2016 IEEE.; 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016 ; Conference date: 17-10-2016 Through 20-10-2016",

year = "2017",

month = may,

day = "2",

doi = "10.1109/ISCSLP.2016.7918436",

language = "English",

series = "Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016",

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

editor = "Hsin-Min Wang and Qingzhi Hou and Yuan Wei and Tan Lee and Jianguo Wei and Lei Xie and Hui Feng and Jianwu Dang and Jianwu Dang",

booktitle = "Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016",

}

Rao, W, Xiao, X, Xu, C, Xu, H, Lee, K, Chng, ES & Li, H 2017, Neural networks based channel compensation for i-vector speaker verification. in H-M Wang, Q Hou, Y Wei, T Lee, J Wei, L Xie, H Feng, J Dang & J Dang (eds), Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016., 7918436, Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016, Institute of Electrical and Electronics Engineers Inc., 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016, Tianjin, China, 17/10/16. https://doi.org/10.1109/ISCSLP.2016.7918436

Neural networks based channel compensation for i-vector speaker verification. / Rao, Wei; Xiao, Xiong; Xu, Chenglin et al.
Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016. ed. / Hsin-Min Wang; Qingzhi Hou; Yuan Wei; Tan Lee; Jianguo Wei; Lei Xie; Hui Feng; Jianwu Dang; Jianwu Dang. Institute of Electrical and Electronics Engineers Inc., 2017. 7918436 (Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016).

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

TY - GEN

T1 - Neural networks based channel compensation for i-vector speaker verification

AU - Rao, Wei

AU - Xiao, Xiong

AU - Xu, Chenglin

AU - Xu, Haihua

AU - Lee, Kongaik

AU - Chng, Eng Siong

AU - Li, Haizhou

PY - 2017/5/2

Y1 - 2017/5/2

N2 - Linear discriminant analysis (LDA) and Gaussian probabilistic LDA (PLDA) have been shown to effectively suppress channel-and session-variability of i-vectors. But they suffer the following limitations: 1) In LDA, a single linear transformation may not be adequate to describe the nonlinear relationship of features and 2) Gaussian-PLDA assumes the speaker and channel factors follow a Gaussian distribution, but they are actually non-Gaussians. We consider neural networks (NN) as a way to overcome the limitations, that captures the nonlinear relationship of features and does not require prior assumptions. This paper investigates three NN based channel compensation methods: deep metric learning, NN classifier, and deep denoising autoencoder and compares their performance with LDA and PLDA. Experiments conducted on NIST 2010 speaker recognition evaluation suggest that NN-based channel compensation methods are superior to LDA and that the performance of NN classifier is better than that of PLDA under most of common evaluation conditions. Additionally, this paper also helps us understand the relationships among LDA, PLDA, and NN based methods.

AB - Linear discriminant analysis (LDA) and Gaussian probabilistic LDA (PLDA) have been shown to effectively suppress channel-and session-variability of i-vectors. But they suffer the following limitations: 1) In LDA, a single linear transformation may not be adequate to describe the nonlinear relationship of features and 2) Gaussian-PLDA assumes the speaker and channel factors follow a Gaussian distribution, but they are actually non-Gaussians. We consider neural networks (NN) as a way to overcome the limitations, that captures the nonlinear relationship of features and does not require prior assumptions. This paper investigates three NN based channel compensation methods: deep metric learning, NN classifier, and deep denoising autoencoder and compares their performance with LDA and PLDA. Experiments conducted on NIST 2010 speaker recognition evaluation suggest that NN-based channel compensation methods are superior to LDA and that the performance of NN classifier is better than that of PLDA under most of common evaluation conditions. Additionally, this paper also helps us understand the relationships among LDA, PLDA, and NN based methods.

KW - I-vector

KW - LDA

KW - Neural networks

KW - PLDA

KW - Speaker recognition

UR - http://www.scopus.com/inward/record.url?scp=85020174813&partnerID=8YFLogxK

U2 - 10.1109/ISCSLP.2016.7918436

DO - 10.1109/ISCSLP.2016.7918436

M3 - Conference article published in proceeding or book

AN - SCOPUS:85020174813

T3 - Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016

BT - Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016

A2 - Wang, Hsin-Min

A2 - Hou, Qingzhi

A2 - Wei, Yuan

A2 - Lee, Tan

A2 - Wei, Jianguo

A2 - Xie, Lei

A2 - Feng, Hui

A2 - Dang, Jianwu

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016

Y2 - 17 October 2016 through 20 October 2016

ER -

Rao W, Xiao X, Xu C, Xu H, Lee K, Chng ES et al. Neural networks based channel compensation for i-vector speaker verification. In Wang HM, Hou Q, Wei Y, Lee T, Wei J, Xie L, Feng H, Dang J, Dang J, editors, Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016. Institute of Electrical and Electronics Engineers Inc. 2017. 7918436. (Proceedings of 2016 10th International Symposium on Chinese Spoken Language Processing, ISCSLP 2016). doi: 10.1109/ISCSLP.2016.7918436

Neural networks based channel compensation for i-vector speaker verification

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this