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 2021-06-19 15:00
Investigation on fine-tuning with image classification networks for deep neural network-based musical instrument classification
Yuki Shiroma, Yuma Kinoshita, Sayaka Shiota, Hitoshi Kiya (TMU) SP2021-17
Abstract (in Japanese) (See Japanese page) 
(in English) In this paper, we investigate abilities of channel conversion methods for fine-tuning with image classification networks under deep neural network-based musical instrument classification. Recently, many deep neural network-based methods have been proposed for scene classification, emotion recognition tasks, and so on. It has also been reported that fine-tuning techniques with well-trained networks using large-scale image dataset improve the performance of sound classification tasks when the limited amount of training data is available. In this case, while a spectrogram extracted from sound data is usually regarded as an image and inputted to the fine-tuned networks with the image classification tasks, the spectrogram image is not suitable to the fine-tuned network because the input of the image classification networks assumes the three channel data like RGB. In this case, the spectrogram is required to be converted to the three channel data, and many methods such as spectrogram duplication method, a method using delta as coefficients and colorization of a spectrogram have been proposed. However, there is no discussion how these methods affect the accuracies. Therefore, we compare various channel conversion methods via fine-tuning of the image classification networks. In the experiments, we performed musical instrument classificaiton with fine-tuning of the well-trained networks by ImageNet. From the results, compared among six channel conversion methods, the colorization of a spectrogram was the most suitable for the fine-tuning with the image classification networks.
Keyword (in Japanese) (See Japanese page) 
(in English) Acoustic musical instrument classification / image classification network / fine-tuning / channel conversion / / / /  
Reference Info. IEICE Tech. Rep., vol. 121, no. 66, SP2021-17, pp. 75-79, June 2021.
Paper # SP2021-17 
Date of Issue 2021-06-11 (SP) 
ISSN Online edition: ISSN 2432-6380
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 SP2021-17

Conference Information
Conference Date 2021-06-18 - 2021-06-19 
Place (in Japanese) (See Japanese page) 
Place (in English) Online 
Topics (in Japanese) (See Japanese page) 
Topics (in English) OTOGAKU Symposium 2021 
Paper Information
Registration To SP 
Conference Code 2021-06-SP-SLP-MUS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) Investigation on fine-tuning with image classification networks for deep neural network-based musical instrument classification 
Sub Title (in English)  
Keyword(1) Acoustic musical instrument classification  
Keyword(2) image classification network  
Keyword(3) fine-tuning  
Keyword(4) channel conversion  
1st Author's Name Yuki Shiroma  
1st Author's Affiliation Tokyo Metropolitan University (TMU)
2nd Author's Name Yuma Kinoshita  
2nd Author's Affiliation Tokyo Metropolitan University (TMU)
3rd Author's Name Sayaka Shiota  
3rd Author's Affiliation Tokyo Metropolitan University (TMU)
4th Author's Name Hitoshi Kiya  
4th Author's Affiliation Tokyo Metropolitan University (TMU)
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 2021-06-19 15:00:00 
Presentation Time 120 minutes 
Registration for SP 
Paper # SP2021-17 
Volume (vol) vol.121 
Number (no) no.66 
Page pp.75-79 
Date of Issue 2021-06-11 (SP) 

[Return to Top Page]

[Return to IEICE Web Page]

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