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 2016-08-25 13:35
Diversity-driven Semi-supervised Ensemble DNN Acoustic Model Training
Sheng Li (Kyoto Univ.), Xugang Lu (NICT), Shinsuke Sakai, Tatsuya Kawahara (Kyoto Univ.) SP2016-40
Abstract (in Japanese) (See Japanese page) 
(in English) We focus on effective training DNN (Deep Neural Network) acoustic models for Chinese spoken lectures with only limited labeled speech and abundant unlabeled speech. Unlike selectively using the unlabeled data in most semi-supervised DNN training methods and working only under supervised setting in previous ensemble DNN training methods, we work on more generalized ensemble training method for both labeled and unlabeled data. In our proposed method, a pair of models is trained in parallel with diverse labels generated for unlabeled data. Together with the standard cross entropy, the KL divergence between each individual model over unlabeled data is incorporated during training. Experiments show that our proposed method can effectively utilize unlabeled data and outperforms other well-established semi-supervised methods.
Keyword (in Japanese) (See Japanese page) 
(in English) Speech recognition / Acoustic model / DNN / Semi-superivsed training / Ensemble training / / /  
Reference Info. IEICE Tech. Rep., vol. 116, no. 189, SP2016-40, pp. 71-76, Aug. 2016.
Paper # SP2016-40 
Date of Issue 2016-08-17 (SP) 
ISSN Print edition: ISSN 0913-5685    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 SP2016-40

Conference Information
Committee SP  
Conference Date 2016-08-24 - 2016-08-25 
Place (in Japanese) (See Japanese page) 
Place (in English) ACCMS, Kyoto Univ. 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Audio event processing, etc. 
Paper Information
Registration To SP 
Conference Code 2016-08-SP 
Language English 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Diversity-driven Semi-supervised Ensemble DNN Acoustic Model Training 
Sub Title (in English)  
Keyword(1) Speech recognition  
Keyword(2) Acoustic model  
Keyword(3) DNN  
Keyword(4) Semi-superivsed training  
Keyword(5) Ensemble training  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Sheng Li  
1st Author's Affiliation Kyoto University (Kyoto Univ.)
2nd Author's Name Xugang Lu  
2nd Author's Affiliation National Institute of Information and Communications Technology (NICT)
3rd Author's Name Shinsuke Sakai  
3rd Author's Affiliation Kyoto University (Kyoto Univ.)
4th Author's Name Tatsuya Kawahara  
4th Author's Affiliation Kyoto University (Kyoto Univ.)
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 ()
Speaker Author-1 
Date Time 2016-08-25 13:35:00 
Presentation Time 25 minutes 
Registration for SP 
Paper # SP2016-40 
Volume (vol) vol.116 
Number (no) no.189 
Page pp.71-76 
#Pages
Date of Issue 2016-08-17 (SP) 


[Return to Top Page]

[Return to IEICE Web Page]


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