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 2023-03-01 09:30
A Study on Scheduled Sampling for Neural Transducer-based ASR
Takafumi Moriya, Takanori Ashihara, Hiroshi Sato, Kohei Matsuura, Tomohiro Tanaka, Ryo Masumura (NTT) EA2022-100 SIP2022-144 SP2022-64
Abstract (in Japanese) (See Japanese page) 
(in English) In this paper, we propose scheduled sampling approaches suited for the recurrent neural network-transducer (RNNT) that is a promising approach for automatic speech recognition (ASR). SS is a technique to train autoregressive model robustly to past errors by randomly replacing some ground-truth tokens with actual outputs generated from a model. SS mitigates the gaps between training and decoding steps, known as exposure bias, and it is often used for attentional encoder-decoder training. However, SS has not been fully examined for RNNT because of the difficulty in applying SS to RNNT due to the complicated RNNT output form. Our SS approaches sample the tokens generated from the distribution of RNNT itself, i.e. internal language model or RNNT outputs. Experiments in three datasets confirm that RNNT trained with our SS approach achieves the best ASR performance. In particular, on a Japanese ASR task, our best system outperforms the previous state-of-the-art alternative.
Keyword (in Japanese) (See Japanese page) 
(in English) automatic speech recognition / neural network / end-to-end / recurrent neural network-transducer / scheduled sampling / / /  
Reference Info. IEICE Tech. Rep., vol. 122, no. 389, SP2022-64, pp. 147-152, Feb. 2023.
Paper # SP2022-64 
Date of Issue 2023-02-21 (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 EA2022-100 SIP2022-144 SP2022-64

Conference Information
Committee SP IPSJ-SLP EA SIP  
Conference Date 2023-02-28 - 2023-03-01 
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 2023-02-SP-SLP-EA-SIP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) A Study on Scheduled Sampling for Neural Transducer-based ASR 
Sub Title (in English)  
Keyword(1) automatic speech recognition  
Keyword(2) neural network  
Keyword(3) end-to-end  
Keyword(4) recurrent neural network-transducer  
Keyword(5) scheduled sampling  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Takafumi Moriya  
1st Author's Affiliation NTT (NTT)
2nd Author's Name Takanori Ashihara  
2nd Author's Affiliation NTT (NTT)
3rd Author's Name Hiroshi Sato  
3rd Author's Affiliation NTT (NTT)
4th Author's Name Kohei Matsuura  
4th Author's Affiliation NTT (NTT)
5th Author's Name Tomohiro Tanaka  
5th Author's Affiliation NTT (NTT)
6th Author's Name Ryo Masumura  
6th Author's Affiliation NTT (NTT)
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 2023-03-01 09:30:00 
Presentation Time 20 minutes 
Registration for SP 
Paper # EA2022-100, SIP2022-144, SP2022-64 
Volume (vol) vol.122 
Number (no) no.387(EA), no.388(SIP), no.389(SP) 
Page pp.147-152 
#Pages 6 
Date of Issue 2023-02-21 (EA, SIP, SP) 

[Return to Top Page]

[Return to IEICE Web Page]
The Institute of Electronics, Information and Communication Engineers (IEICE), Japan