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 2021-03-04 09:00
Anomalous Sound Detection Using a Binary Classification Model Considering Class Centroids
Ibuki Kuroyanagi, Tomiki Hayashi, Kazuya Takeda, Tomoki Toda (Nagoya Univ) EA2020-79 SIP2020-110 SP2020-44
Abstract (in Japanese) (See Japanese page) 
(in English) In an anomalous sound detection system, it is necessary to detect unknown anomalous sounds using only normal sound data.
On the other hand, considering the anomalous sound detection system's operation, it is desirable to build a system that can also use a small amount of anomalous sound data accumulated through the operation.
Therefore, we focus on a binary classification model that can easily use anomalous sound data as well as normal sound data while further using outlier data belonging to the other domain as pseudo-anomalous sound data.
This report presents a new loss function based on metric learning for this binary classification, making it possible to learn the distance relationship from each class centroid in the feature space.
The multi-task learning of classification and metric learning enables us to learn a mapping to the feature space where the within-class variance is minimized and the between-class variance is maximized while keeping normal and anomalous classes linearly separable.
We also investigate the relationship between the amount of anomalous sound data used for training and the detection performance to clarify that the additional use of anomalous sound data is effective for further improving the binary classification model.
Through the experimental evaluation with DCASE~2020 Task~2 dataset, we demonstrate that 1) multi-task learning of binary classification and metric learning is effective, 2) the loss function considering the distance from the class centroids in the feature space is also effective, and 3) using a small amount of anomalous sound data significantly improves the detection performance.
Keyword (in Japanese) (See Japanese page) 
(in English) anomalous sound detection / binary classification / class centriods / semi-supervised learning / metric learning / multi task learning / /  
Reference Info. IEICE Tech. Rep., vol. 120, no. 397, EA2020-79, pp. 114-121, March 2021.
Paper # EA2020-79 
Date of Issue 2021-02-24 (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 EA2020-79 SIP2020-110 SP2020-44

Conference Information
Committee EA US SP SIP IPSJ-SLP  
Conference Date 2021-03-03 - 2021-03-04 
Place (in Japanese) (See Japanese page) 
Place (in English) Online 
Topics (in Japanese) (See Japanese page) 
Topics (in English) Speech, Engineering/Electro Acoustics, Signal Processing, Ultrasonics, and Related Topics 
Paper Information
Registration To EA 
Conference Code 2021-03-EA-US-SP-SIP-SLP 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Anomalous Sound Detection Using a Binary Classification Model Considering Class Centroids 
Sub Title (in English)  
Keyword(1) anomalous sound detection  
Keyword(2) binary classification  
Keyword(3) class centriods  
Keyword(4) semi-supervised learning  
Keyword(5) metric learning  
Keyword(6) multi task learning  
Keyword(7)  
Keyword(8)  
1st Author's Name Ibuki Kuroyanagi  
1st Author's Affiliation Nagoya University (Nagoya Univ)
2nd Author's Name Tomiki Hayashi  
2nd Author's Affiliation Nagoya University (Nagoya Univ)
3rd Author's Name Kazuya Takeda  
3rd Author's Affiliation Nagoya University (Nagoya Univ)
4th Author's Name Tomoki Toda  
4th Author's Affiliation Nagoya University (Nagoya 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 2021-03-04 09:00:00 
Presentation Time 25 minutes 
Registration for EA 
Paper # EA2020-79, SIP2020-110, SP2020-44 
Volume (vol) vol.120 
Number (no) no.397(EA), no.398(SIP), no.399(SP) 
Page pp.114-121 
#Pages
Date of Issue 2021-02-24 (EA, SIP, SP) 


[Return to Top Page]

[Return to IEICE Web Page]


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