Speaker embedding extraction with phonetic information

Yi Liu; Liang He; Jia Liu; Michael T. Johnson

doi:10.21437/Interspeech.2018-1226

Speaker embedding extraction with phonetic information

Yi Liu, Liang He, Jia Liu, Michael T. Johnson

Research output: Contribution to journal › Conference article › peer-review

33 Scopus citations

Abstract

Speaker embeddings achieve promising results on many speaker verification tasks. Phonetic information, as an important component of speech, is rarely considered in the extraction of speaker embeddings. In this paper, we introduce phonetic information to the speaker embedding extraction based on the x-vector architecture. Two methods using phonetic vectors and multi-task learning are proposed. On the Fisher dataset, our best system outperforms the original x-vector approach by 20% in EER, and by 15%, 15% in minDCF08 and minDCF10, respectively. Experiments conducted on NIST SRE10 further demonstrate the effectiveness of the proposed methods.

Original language	English
Pages (from-to)	2247-2251
Number of pages	5
Journal	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	2018-September
DOIs	https://doi.org/10.21437/Interspeech.2018-1226
State	Published - 2018
Event	19th Annual Conference of the International Speech Communication, INTERSPEECH 2018 - Hyderabad, India Duration: Sep 2 2018 → Sep 6 2018

Bibliographical note

Publisher Copyright:
© 2018 International Speech Communication Association. All rights reserved.

Funding

The work is supported by National Natural Science Foundation of China under Grant No. 61370034, No. 61403224 and No. 61273268.

Funders	Funder number
National Natural Science Foundation of China (NSFC)	61403224, 61273268, 61370034

Keywords

Multi-task learning
Phonetic information
Phonetic vectors
Speaker embedding
Speaker verification

ASJC Scopus subject areas

Language and Linguistics
Human-Computer Interaction
Signal Processing
Software
Modeling and Simulation

Access to Document

10.21437/Interspeech.2018-1226

Cite this

@article{257dfc26000d40e897b558bc85183b94,

title = "Speaker embedding extraction with phonetic information",

abstract = "Speaker embeddings achieve promising results on many speaker verification tasks. Phonetic information, as an important component of speech, is rarely considered in the extraction of speaker embeddings. In this paper, we introduce phonetic information to the speaker embedding extraction based on the x-vector architecture. Two methods using phonetic vectors and multi-task learning are proposed. On the Fisher dataset, our best system outperforms the original x-vector approach by 20% in EER, and by 15%, 15% in minDCF08 and minDCF10, respectively. Experiments conducted on NIST SRE10 further demonstrate the effectiveness of the proposed methods.",

keywords = "Multi-task learning, Phonetic information, Phonetic vectors, Speaker embedding, Speaker verification",

author = "Yi Liu and Liang He and Jia Liu and Johnson, {Michael T.}",

note = "Publisher Copyright: {\textcopyright} 2018 International Speech Communication Association. All rights reserved.; 19th Annual Conference of the International Speech Communication, INTERSPEECH 2018 ; Conference date: 02-09-2018 Through 06-09-2018",

year = "2018",

doi = "10.21437/Interspeech.2018-1226",

language = "English",

volume = "2018-September",

pages = "2247--2251",

}

TY - JOUR

T1 - Speaker embedding extraction with phonetic information

AU - Liu, Yi

AU - He, Liang

AU - Liu, Jia

AU - Johnson, Michael T.

PY - 2018

Y1 - 2018

N2 - Speaker embeddings achieve promising results on many speaker verification tasks. Phonetic information, as an important component of speech, is rarely considered in the extraction of speaker embeddings. In this paper, we introduce phonetic information to the speaker embedding extraction based on the x-vector architecture. Two methods using phonetic vectors and multi-task learning are proposed. On the Fisher dataset, our best system outperforms the original x-vector approach by 20% in EER, and by 15%, 15% in minDCF08 and minDCF10, respectively. Experiments conducted on NIST SRE10 further demonstrate the effectiveness of the proposed methods.

AB - Speaker embeddings achieve promising results on many speaker verification tasks. Phonetic information, as an important component of speech, is rarely considered in the extraction of speaker embeddings. In this paper, we introduce phonetic information to the speaker embedding extraction based on the x-vector architecture. Two methods using phonetic vectors and multi-task learning are proposed. On the Fisher dataset, our best system outperforms the original x-vector approach by 20% in EER, and by 15%, 15% in minDCF08 and minDCF10, respectively. Experiments conducted on NIST SRE10 further demonstrate the effectiveness of the proposed methods.

KW - Multi-task learning

KW - Phonetic information

KW - Phonetic vectors

KW - Speaker embedding

KW - Speaker verification

UR - http://www.scopus.com/inward/record.url?scp=85054953560&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85054953560&partnerID=8YFLogxK

U2 - 10.21437/Interspeech.2018-1226

DO - 10.21437/Interspeech.2018-1226

M3 - Conference article

AN - SCOPUS:85054953560

SN - 2308-457X

VL - 2018-September

SP - 2247

EP - 2251

JO - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

JF - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

T2 - 19th Annual Conference of the International Speech Communication, INTERSPEECH 2018

Y2 - 2 September 2018 through 6 September 2018

ER -

Speaker embedding extraction with phonetic information

Abstract

Bibliographical note

Funding

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this