Information: Join today and make your research activities more affordable! Technical workshop participation fees and annual registration fees are available at member rates.
Notice: [Important] Announcement of Changes to Registration Fee Payment and Manuscript Upload Procedures for IEICE Technical Meetings
IEICE Technical Committee Submission System
Conference Paper's Information
Online Proceedings
[Sign in]
Tech. Rep. Archives
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2024-02-29 09:50
Pseudo-speaker augmentation based on vocal tract length perturbation considering speaker variability for speaker verification
Fumika Ono, Tomoka Wakamatsu, Sayaka Shiota (TMU) EA2023-62 SIP2023-109 SP2023-44
Abstract (in Japanese) (See Japanese page) 
(in English) In order to construct a reliable speaker verification system based on speaker embeddings, it is necessary to train the speaker embedding extraction model using large-scale training data. Data augmentation is one method for preparing a sufficient amount of training data. Data augmentation in conventional speaker verification mainly involves increasing the number of utterances. However, in the speaker verification systems based on speaker embedding, providing sufficient speakers is also important for improving accuracy. As a method for increasing the number of speakers, the effectiveness of generating pseudo-speakers using vocal tract length perturbations (VTLP) has also been reported. On the other hand, there were problems, such as how to set appropriate parameters when generating pseudo-speakers and the number of effective pseudo-speakers. Therefore, in this paper, by automatically adjusting the parameters of VTLP in consideration of the amount of variation in speaker characteristics, we can increase the number of speakers by selecting more pseudo-speakers than in the conventional pseudo-speaker selection method. In the experiment, we generated pseudo-speakers using the proposed method and evaluated it using a speaker verification system based on ECAPA-TDNN, a state-of-the-art method. From the experimental results, we found that in data augmentation for speaker verification based on speaker embedding, we can improve the speaker verification system by using both augmentations of the number of utterances per speaker and the number of speakers that take into account the variation in speaker characteristics.
Keyword (in Japanese) (See Japanese page) 
(in English) speaker verification / pseudo-speaker augmentation / VTLP / ECAPA-TDNN / / / /  
Reference Info. IEICE Tech. Rep., vol. 123, no. 403, SP2023-44, pp. 7-12, Feb. 2024.
Paper # SP2023-44 
Date of Issue 2024-02-22 (EA, SIP, SP) 
ISSN Online edition: ISSN 2432-6380
Copyright
and
reproduction
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034)
Download PDF EA2023-62 SIP2023-109 SP2023-44

Conference Information
Committee SIP SP EA IPSJ-SLP  
Conference Date 2024-02-29 - 2024-03-01 
Place (in Japanese) (See Japanese page) 
Place (in English)  
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To SP 
Conference Code 2024-02-SIP-SP-EA-SLP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Pseudo-speaker augmentation based on vocal tract length perturbation considering speaker variability for speaker verification 
Sub Title (in English)  
Keyword(1) speaker verification  
Keyword(2) pseudo-speaker augmentation  
Keyword(3) VTLP  
Keyword(4) ECAPA-TDNN  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Fumika Ono  
1st Author's Affiliation Tokyo Metropolitan University (TMU)
2nd Author's Name Tomoka Wakamatsu  
2nd Author's Affiliation Tokyo Metropolitan University (TMU)
3rd Author's Name Sayaka Shiota  
3rd Author's Affiliation Tokyo Metropolitan University (TMU)
4th Author's Name  
4th Author's Affiliation ()
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
21st Author's Name  
21st Author's Affiliation ()
22nd Author's Name  
22nd Author's Affiliation ()
23rd Author's Name  
23rd Author's Affiliation ()
24th Author's Name  
24th Author's Affiliation ()
25th Author's Name  
25th Author's Affiliation ()
26th Author's Name / /
26th Author's Affiliation ()
()
27th Author's Name / /
27th Author's Affiliation ()
()
28th Author's Name / /
28th Author's Affiliation ()
()
29th Author's Name / /
29th Author's Affiliation ()
()
30th Author's Name / /
30th Author's Affiliation ()
()
31st Author's Name / /
31st Author's Affiliation ()
()
32nd Author's Name / /
32nd Author's Affiliation ()
()
33rd Author's Name / /
33rd Author's Affiliation ()
()
34th Author's Name / /
34th Author's Affiliation ()
()
35th Author's Name / /
35th Author's Affiliation ()
()
36th Author's Name / /
36th Author's Affiliation ()
()
Speaker Author-1 
Date Time 2024-02-29 09:50:00 
Presentation Time 20 minutes 
Registration for SP 
Paper # EA2023-62, SIP2023-109, SP2023-44 
Volume (vol) vol.123 
Number (no) no.401(EA), no.402(SIP), no.403(SP) 
Page pp.7-12 
#Pages
Date of Issue 2024-02-22 (EA, SIP, SP) 


[Return to Top Page]

[Return to IEICE Web Page]


The Institute of Electronics, Information and Communication Engineers (IEICE), Japan