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 2026-06-06 13:00
[Poster Presentation] Visual Speech Recognition via Large Language Model Distillation and Viseme-Aware CTC Learning
Haruki Komai, Koichi Shinoda (Science Tokyo) SP2026-17
Abstract (in Japanese) (See Japanese page) 
(in English) Using a large language model (LLM) as a decoder for visual speech recognition often incurs substantial inference cost. This paper investigates VSR-KD-CTC, a low-cost VSR framework that transfers the linguistic knowledge of an LLM to a lightweight decoder through knowledge distillation. The proposed method combines the standard cross-entropy loss with a distillation loss that aligns the decoder output distribution with that of the teacher LLM. As a result, inference is performed using only a lightweight encoder-decoder architecture without executing the LLM. Experiments on LRS2 show that, compared with VSP-LLM, the proposed method reduces the inference cost from 1693 GFLOPs to 126 GFLOPs and decreases the total parameter count from 7251M to 509M. An ablation study on the same lightweight VSR configuration further shows that LLM distillation improves the WER from 25.9% to 25.4%, indicating that the cost reduction does not require sacrificing recognition accuracy. In contrast, viseme-based CTC auxiliary learning does not provide additional gains under the present experimental conditions.
Keyword (in Japanese) (See Japanese page) 
(in English) visual speech recognition / large language model / knowledge distillation / Viseme / CTC / / /  
Reference Info. IEICE Tech. Rep., vol. 126, no. 58, SP2026-17, pp. 91-98, June 2026.
Paper # SP2026-17 
Date of Issue 2026-05-29 (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 SP2026-17

Conference Information
Committee SP IPSJ-SLP IPSJ-MUS  
Conference Date 2026-06-05 - 2026-06-06 
Place (in Japanese) (See Japanese page) 
Place (in English) The University of Electro-Communications 
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To SP 
Conference Code 2026-06-SP-SLP-MUS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Visual Speech Recognition via Large Language Model Distillation and Viseme-Aware CTC Learning 
Sub Title (in English)  
Keyword(1) visual speech recognition  
Keyword(2) large language model  
Keyword(3) knowledge distillation  
Keyword(4) Viseme  
Keyword(5) CTC  
Keyword(6)  
Keyword(7)  
Keyword(8)  
1st Author's Name Haruki Komai  
1st Author's Affiliation Institute of Science Tokyo (Science Tokyo)
2nd Author's Name Koichi Shinoda  
2nd Author's Affiliation Institute of Science Tokyo (Science Tokyo)
3rd Author's Name  
3rd Author's Affiliation ()
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 2026-06-06 13:00:00 
Presentation Time 180 minutes 
Registration for SP 
Paper # SP2026-17 
Volume (vol) vol.126 
Number (no) no.58 
Page pp.91-98 
#Pages 8 
Date of Issue 2026-05-29 (SP) 

[Return to Top Page]

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