基于面部行为与语音融合特征的重度抑郁识别

北京邮电大学学报 ›› 2023, Vol. 46 ›› Issue (1): 32-37.

基于面部行为与语音融合特征的重度抑郁识别

李锦珑^1,2,陈琼琼³,丁志杰⁴,刘振宇¹

1. 兰州大学 2.兰州工业学院 3. 甘肃省第二人民医院 4. 天水市第三人民医院

收稿日期:2022-01-14 修回日期:2022-03-12 出版日期:2023-02-28 发布日期:2023-02-22
通讯作者: 刘振宇 E-mail:liuzhenyu@lzu.edu.cn
基金资助:
国家重点研发计划项目; 甘肃省科技计划项目

Recognition of Major Depressive Disorder Based on Facial Behavior and Speech Fusion Features

Received:2022-01-14 Revised:2022-03-12 Online:2023-02-28 Published:2023-02-22

摘要/Abstract

摘要： 针对实现基于计算机辅助的自动抑郁识别的迫切需求,开展了面向中国人群的基于面部行为和语音特征的重度抑郁识别研究。首先,从已构建的抑郁症数据集中选取符合研究条件的 72 个样本进行分析;然后,采用变分模态分解法将面部活动和语音数据分解为不同的频率分量,并通过分析不同频段的能量分布来构建特征集; 最后,采用多种分类器的投票决策实现抑郁识别。实验结果表明,男性被试组的识别正确率可达到 81.1% ,女性被试组可达到 78.7% ,在保护被试个人隐私的前提下获得了较高的分类结果。

关键词: 抑郁识别, 多模态融合, 面部行为, 语音信号, 变分模态分解

Abstract: In response to the urgent need to realize computer-assisted automatic depression recognition, a major depressive disorder recognition study based on facial behavior and speech features for the Chinese population has been carried out. First, 72 samples conforming to the research conditions are selected from the constructed depression data set for analysis. Then, the facial activity and speech data are decomposed into different frequency components by using the variational modal decomposition method, and the feature set is constructed by analyzing the energy distribution of different frequency bands. Finally, depression recognition is achieved by using the voting decisions of several classifiers. The experimental results show that the recognition accuracy of the male group can reach 81.1% , and the recognition accuracy of the female group can reach 78.7% . Under the premise of protecting the privacy of the subjects, higher
classification results are obtained.

Key words: depression recognition , multimodal fusion , facial behavior , speech signal , variational mode decomposition

中图分类号:

TP391.4

李锦珑陈琼琼丁志杰刘振宇. 基于面部行为与语音融合特征的重度抑郁识别[J]. 北京邮电大学学报, 2023, 46(1): 32-37.

LI Jinlong, CHEN Qiongqiong, DING Zhijie, LIU Zhenyu. Recognition of Major Depressive Disorder Based on Facial Behavior and Speech Fusion Features[J]. Journal of Beijing University of Posts and Telecommunications, 2023, 46(1): 32-37.