JAIST Repository >
School of Information Science >
Conference Papers >
Conference Papers >

Please use this identifier to cite or link to this item: http://hdl.handle.net/10119/15776

Title: Maximal Information Coefficient and Predominant Correlation-Based Feature Selection Toward A Three-Layer Model for Speech Emotion Recognition
Authors: Li, Xingfeng
Akagi, Masato
Issue Date: 2018-11-15
Publisher: Institute of Electrical and Electronics Engineers (IEEE)
Magazine name: 2018 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)
Start page: 1428
End page: 1434
DOI: 10.23919/APSIPA.2018.8659695
Abstract: This paper describes an efficient attempt to build a three-layer emotion perception model consisting of acoustic features, semantic primitives, and emotion dimensions with a focus on acoustic feature subset selection. Previous studies using this model focused on the most relevant acoustic features using a Pearson correlation coefficient-based filter approach, which could only capture the relation limited to linear function well. However, perception of human emotion is vague; linear correlation measures could not capture the relations that are not linear in nature. In this study, we introduce a novel feature selection algorithm based on the maximal information coefficient and predominant correlation, which can identify relevant features between paired variables in spite of linear or nonlinear relations and remove redundancies among the relevant features. Experimental results on the Berlin Emo-DB and Chinese Emotional Speech Corpus demonstrated that the proposed algorithm achieves an improvement to estimation of emotion dimensions, resulting in a smaller mean absolute error and higher correlation coefficient between estimations and human evaluations, compared with the referred Pearson correlation coefficient-based method, and the commonly used wrapper-based method of sequential floating forward selection.
Rights: This is the author's version of the work. Copyright (C) 2018 IEEE. 2018 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), 2018, 1428-1434. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
URI: http://hdl.handle.net/10119/15776
Material Type: author
Appears in Collections:b11-1. 会議発表論文・発表資料 (Conference Papers)

Files in This Item:

File Description SizeFormat
2907.pdf550KbAdobe PDFView/Open

All items in DSpace are protected by copyright, with all rights reserved.

 


Contact : Library Information Section, Japan Advanced Institute of Science and Technology