基于语音脑电的双模态心理压力分级评估研究
Evaluation of psychological stress level based on speech and EEG signal
杜扶遥 1姜囡 2陆思宇1
作者信息
- 1. 中国刑事警察学院公安信息技术与情报学院 沈阳 110854
- 2. 中国刑事警察学院公安信息技术与情报学院 沈阳 110854;证据科学教育部重点实验室(中国政法大学) 北京 100088
- 折叠
摘要
为了有效提高压力分级方法的精确度,实现多模态信息交互和多维立体融合特征的深层挖掘,提出一种基于模型分级的多模态压力识别方法.基于语音信号振幅特征和脑电信号各频段波幅特征,构建新的心理压力指数模型,并提出针对该模型的心理压力分级方法,有效解决了主观评估精度受限以及压力分类依据不明确等问题.以模型分级为依据重制 MAHNOB-HCI数据集标签,构建了包含脑电时频空信息和语音时频信息的立体多维融合特征,避免了单特征识别方法导致的压力信息缺失问题.与单模态识别方法的对比分析,本文提出方法识别准确率分别提高了10.72%和3.36%;与常规双模态方法的对比分析,识别准确率提高了7.51%.综上表明,本文所提方法能够更准确的揭示异构数据全频段信息与心理压力的关联关系,有效提升了识别性能.
Abstract
In order to improve the accuracy of pressure classification method.realize the deep mining of multi-modal information interaction and multi-dimensional three-dimensional fusion features,a multi-modal pressure identification method based on model classification is proposed.A new psychological stress index model is constructed based on the amplitude characteristics of speech signals and the amplitude characteristics of each frequency band of EEG signals,and a psychological stress classification method for the model is proposed to solve the problems of limited subjective assessment accuracy and unclear stress classification basis.The labels of MAHNOB-HCI data set are reconstructed based on the model classification,and the multi-dimensional stereo fusion features containing EEG time-frequency-space information and speech time-frequency information are constructed to solve the problem of missing pressure information caused by the single feature research method.Compared with the single modal method,the recognition accuracy of the proposed method is increased by 10.72%and 3.36%,respectively.Compared with the conventional dual-modal method,the recognition accuracy is increased by 7.51%.To sum up,the proposed method can more accurately reveal the relationship between the full-band information of heterogeneous data and psychological stress,and effectively improve the recognition performance.
关键词
脑电信号/语音信号/双模态/心理压力分级/多维融合特征Key words
EEG signals/voice signal/bimodal/psychological stress classification/multi-dimensional fusion feature引用本文复制引用
出版年
2024
NETL
NSTL
万方数据