基于深度学习进行多平台语音交互的抑郁症监测系统  

Depression Monitoring System Based on Deep Learning for Multi-platform Speech Interaction

作　　者：吴伟 王琦翔 马龙华 赵祥红[2] WU Wei;WANG Qi-xiang;MA Long-hua;ZHAO Xiang-hong(Zhejiang Sci-Tech University,Hangzhou 310018,China;Ningbo Tech University,Ningbo 315100,China)

机构地区：[1]浙江理工大学,浙江杭州310018 [2]浙大宁波理工学院,浙江宁波315100

出　　处：《浙江工商职业技术学院学报》2023年第3期45-50,共6页Journal of Zhejiang Business Technology Institute

基　　金：国家自然科学基金项目“大数据环境下的用户隐私保护与系统风险控制的融合问题研究”(编号61972350),宁波市自然科学基金项目“一种多模态隐式可穿戴心理压力监测方法”(编号2022J165)阶段性研究成果。

摘　　要：针对目前主流的抑郁诊断方法存在主观性较强或成本较高的问题,设计了一个多平台语音收集和分析的抑郁症监测系统,并提出一种针对抑郁症检测的神经网络模型。此系统通过网页、小程序和手机应用程序收集语音信号,然后上传到云端服务器,最后将预测结果反馈到客户端显示。除了以上三种不同的渠道采集语音以外,该系统可以扩展其他收集语音的方法。并且也将提出的神经网络模型与另外五种经典的模型进行比较,提出的模型在公开的抑郁症语音数据的测试集上平均准确率达到89.6%,相比于支持向量机、线性判别算法、K邻近算法、随机森林算法和逻辑回归算法分别提高15.25%、14.31%、10.39%、13.23%和14.71%。将训练好的模型部署到服务器上,78个参与者使用该系统后,统计预测的平均准确率达到82.6%。因此,该系统是一种低成本、客观且非侵入的精神诊断方案,可以提高抑郁症的诊断准确率。A depression monitoring system was designed to address the subjectivity and high cost issues of current mainstream depression diagnostic methods.The system utilizes multi-platform speech collection and analysis and proposes a neural network model for depression detection.Speech signals are collected through web pages,mini programs and mobile applications and uploaded to cloud servers.The prediction results are then fed back to the client for display.In addition to these three different channels for collecting speeches,the system can also expand to other methods of speech collection.The proposed neural network model was compared with five other classical models and achieved an average accuracy of 89.6%on the test set of publicly available depression speech data,which improved by 15.25%,14.31%,10.39%,13.23%and 14.71%compared to support vector machine,linear discriminant algorithm,K-nearest neighbor algorithm,random forest algorithm and logistic regression algorithm respectively.After the trained model was deployed to the server,the average prediction accuracy of 78 participants using the system reached 82.6%.Therefore,the system is a low-cost,objective and non-invasive mental diagnosis solution that can improve the accuracy of depression diagnosis.

关 键 词：抑郁症检测 语音 神经网络模型 低成本 精神诊断方案 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]