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 2025-03-02 09:30
Uncertainty-Based Streaming ASR with Evidential Deep Learning
Hiroaki Sato, Asahi Sakuma, Ryuga Sugano, Tadashi Kumano, Yoshihiko Kawai (NHK STRL), Ogawa Tetsuji (Waseda Univ.) EA2024-77 SIP2024-112 SP2024-18
Abstract (in Japanese) (See Japanese page) 
(in English) We propose an attention-based encoder-decoder (AED) speech recognition model for streaming recognition using the evidential deep learning (EDL) framework. While AED models achieve high accuracy in offline speech recognition, they struggle with streaming recognition, outputting incorrect tokens before sufficient input is received. EDL models uncertainty using the Dirichlet distribution, addressing the limitation of softmax-based methods that do not account for output uncertainty. We introduce EDL into AED models to control token output timing based on uncertainty. During training, we use loss functions to decrease model uncertainty as more speech input is received. During inference, token output decisions are made by thresholding uncertainty values. Experiments on CSJ and Librispeech show that our method achieves superior streaming performance compared to existing approaches without modifying the AED model structure. We also show that error rate remains stable with varying speech input lengths, and that error rate and latency can be controlled by adjusting the threshold.
Keyword (in Japanese) (See Japanese page) 
(in English) streaming ASR / evidential deep learning / attention-based encoder-decoder / / / / /  
Reference Info. IEICE Tech. Rep., vol. 124, no. 391, SP2024-18, pp. 1-6, March 2025.
Paper # SP2024-18 
Date of Issue 2025-02-23 (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 EA2024-77 SIP2024-112 SP2024-18

Conference Information
Committee EA SIP SP IPSJ-SLP  
Conference Date 2025-03-02 - 2025-03-04 
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 2025-03-EA-SIP-SP-SLP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Uncertainty-Based Streaming ASR with Evidential Deep Learning 
Sub Title (in English)  
Keyword(1) streaming ASR  
Keyword(2) evidential deep learning  
Keyword(3) attention-based encoder-decoder  
Keyword(4)  
Keyword(5)  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Hiroaki Sato  
1st Author's Affiliation NHK Science & Technology Research Laboratories (NHK STRL)
2nd Author's Name Asahi Sakuma  
2nd Author's Affiliation NHK Science & Technology Research Laboratories (NHK STRL)
3rd Author's Name Ryuga Sugano  
3rd Author's Affiliation NHK Science & Technology Research Laboratories (NHK STRL)
4th Author's Name Tadashi Kumano  
4th Author's Affiliation NHK Science & Technology Research Laboratories (NHK STRL)
5th Author's Name Yoshihiko Kawai  
5th Author's Affiliation NHK Science & Technology Research Laboratories (NHK STRL)
6th Author's Name Ogawa Tetsuji  
6th Author's Affiliation Waseda University (Waseda Univ.)
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 2025-03-02 09:30:00 
Presentation Time 20 minutes 
Registration for SP 
Paper # EA2024-77, SIP2024-112, SP2024-18 
Volume (vol) vol.124 
Number (no) no.389(EA), no.390(SIP), no.391(SP) 
Page pp.1-6 
#Pages 6 
Date of Issue 2025-02-23 (EA, SIP, SP) 

[Return to Top Page]

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