Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks

Kwok Kwong Yiu; Man Wai Mak; Sun Yuan Kung

doi:10.1109/ICONIP.2002.1201921

Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks

Kwok Kwong Yiu, Man Wai Mak, Sun Yuan Kung

The Hong Kong Polytechnic University

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

2 Citations (Scopus)

Abstract

This paper compares kernel-based probabilistic neural networks for speaker verification. Experimental evaluations based on 138 speakers of the YOHO corpus using probabilistic decision-based neural networks (PDBNNs), Gaussian mixture models (GMMs) and elliptical basis function networks (EBFNs) as speaker models were conducted. The original PDBNN training algorithm was also modified to make PDBNNs appropriate for speaker verification. Results show that the equal error rate obtained by PDBNNs and GMMs is about half of that of EBFNs (1.19% vs. 2.73%), suggesting that GMM- and PDBNN-based speaker models outperform the EBFN one. This work also finds that the globally supervised learning of PDBNNs is able to find a set of decision thresholds that reduce the variation in FAR, whereas the ad hoc approach used by the EBFNs and GMMs is not able to do so. This property makes the performance of PDBNN-based systems more predictable.

Original language	English
Title of host publication	ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing
Subtitle of host publication	Computational Intelligence for the E-Age
Publisher	IEEE
Pages	2386-2390
Number of pages	5
Volume	5
ISBN (Electronic)	9789810475246, 9810475241
DOIs	https://doi.org/10.1109/ICONIP.2002.1201921
Publication status	Published - 1 Jan 2002
Event	9th International Conference on Neural Information Processing, ICONIP 2002 - Orchid Country Club, Singapore, Singapore Duration: 18 Nov 2002 → 22 Nov 2002

Conference

Conference	9th International Conference on Neural Information Processing, ICONIP 2002
Country/Territory	Singapore
City	Singapore
Period	18/11/02 → 22/11/02

ASJC Scopus subject areas

Computer Networks and Communications
Information Systems
Signal Processing

More information

10.1109/ICONIP.2002.1201921

Cite this

Yiu, K. K., Mak, M. W., & Kung, S. Y. (2002). Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks. In ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing: Computational Intelligence for the E-Age (Vol. 5, pp. 2386-2390). Article 1201921 IEEE. https://doi.org/10.1109/ICONIP.2002.1201921

@inproceedings{97e694aa2375428bbd2103afbd33f28a,

title = "Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks",

abstract = "This paper compares kernel-based probabilistic neural networks for speaker verification. Experimental evaluations based on 138 speakers of the YOHO corpus using probabilistic decision-based neural networks (PDBNNs), Gaussian mixture models (GMMs) and elliptical basis function networks (EBFNs) as speaker models were conducted. The original PDBNN training algorithm was also modified to make PDBNNs appropriate for speaker verification. Results show that the equal error rate obtained by PDBNNs and GMMs is about half of that of EBFNs (1.19% vs. 2.73%), suggesting that GMM- and PDBNN-based speaker models outperform the EBFN one. This work also finds that the globally supervised learning of PDBNNs is able to find a set of decision thresholds that reduce the variation in FAR, whereas the ad hoc approach used by the EBFNs and GMMs is not able to do so. This property makes the performance of PDBNN-based systems more predictable.",

author = "Yiu, {Kwok Kwong} and Mak, {Man Wai} and Kung, {Sun Yuan}",

year = "2002",

month = jan,

day = "1",

doi = "10.1109/ICONIP.2002.1201921",

language = "English",

volume = "5",

pages = "2386--2390",

booktitle = "ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing",

publisher = "IEEE",

address = "United States",

note = "9th International Conference on Neural Information Processing, ICONIP 2002 ; Conference date: 18-11-2002 Through 22-11-2002",

}

Yiu, KK, Mak, MW & Kung, SY 2002, Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks. in ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing: Computational Intelligence for the E-Age. vol. 5, 1201921, IEEE, pp. 2386-2390, 9th International Conference on Neural Information Processing, ICONIP 2002, Singapore, Singapore, 18/11/02. https://doi.org/10.1109/ICONIP.2002.1201921

Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks. / Yiu, Kwok Kwong; Mak, Man Wai; Kung, Sun Yuan.
ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing: Computational Intelligence for the E-Age. Vol. 5 IEEE, 2002. p. 2386-2390 1201921.

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

TY - GEN

T1 - Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks

AU - Yiu, Kwok Kwong

AU - Mak, Man Wai

AU - Kung, Sun Yuan

PY - 2002/1/1

Y1 - 2002/1/1

N2 - This paper compares kernel-based probabilistic neural networks for speaker verification. Experimental evaluations based on 138 speakers of the YOHO corpus using probabilistic decision-based neural networks (PDBNNs), Gaussian mixture models (GMMs) and elliptical basis function networks (EBFNs) as speaker models were conducted. The original PDBNN training algorithm was also modified to make PDBNNs appropriate for speaker verification. Results show that the equal error rate obtained by PDBNNs and GMMs is about half of that of EBFNs (1.19% vs. 2.73%), suggesting that GMM- and PDBNN-based speaker models outperform the EBFN one. This work also finds that the globally supervised learning of PDBNNs is able to find a set of decision thresholds that reduce the variation in FAR, whereas the ad hoc approach used by the EBFNs and GMMs is not able to do so. This property makes the performance of PDBNN-based systems more predictable.

AB - This paper compares kernel-based probabilistic neural networks for speaker verification. Experimental evaluations based on 138 speakers of the YOHO corpus using probabilistic decision-based neural networks (PDBNNs), Gaussian mixture models (GMMs) and elliptical basis function networks (EBFNs) as speaker models were conducted. The original PDBNN training algorithm was also modified to make PDBNNs appropriate for speaker verification. Results show that the equal error rate obtained by PDBNNs and GMMs is about half of that of EBFNs (1.19% vs. 2.73%), suggesting that GMM- and PDBNN-based speaker models outperform the EBFN one. This work also finds that the globally supervised learning of PDBNNs is able to find a set of decision thresholds that reduce the variation in FAR, whereas the ad hoc approach used by the EBFNs and GMMs is not able to do so. This property makes the performance of PDBNN-based systems more predictable.

UR - http://www.scopus.com/inward/record.url?scp=79960384492&partnerID=8YFLogxK

U2 - 10.1109/ICONIP.2002.1201921

DO - 10.1109/ICONIP.2002.1201921

M3 - Conference article published in proceeding or book

VL - 5

SP - 2386

EP - 2390

BT - ICONIP 2002 - Proceedings of the 9th International Conference on Neural Information Processing

PB - IEEE

T2 - 9th International Conference on Neural Information Processing, ICONIP 2002

Y2 - 18 November 2002 through 22 November 2002

ER -

Speaker verification with a priori threshold determination using kernel-based probabilistic neural networks

Abstract

Conference

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this