基于跨用户语音域适应网络的抑郁症检测

doi:10.16180/j.cnki.issn1007-7820.2025.01.012

电子科技 ›› 2025, Vol. 38 ›› Issue (1): 88-94.doi: 10.16180/j.cnki.issn1007-7820.2025.01.012

• • 上一篇

基于跨用户语音域适应网络的抑郁症检测

吴伟¹, 马龙华¹^,², 赵祥红²()

1.浙江理工大学信息科学与工程学院,浙江杭州 310018
2.浙大宁波理工学院信息科学与工程学院,浙江宁波 315100

收稿日期:2023-06-25 修回日期:2023-07-16 出版日期:2025-01-15 发布日期:2025-01-06
通讯作者: 赵祥红(1981-),男, E-mail:wuwei_ah@163.com,博士,副教授。研究方向:机器学习、生物医学信号处理。
作者简介:吴伟(1999-),男,硕士研究生。研究方向:机器学习、嵌入式开发。
马龙华(1965-),男,博士,教授。研究方向:复杂系统综合集成建模。
基金资助:
国家自然科学基金(61972350);国家自然科学基金(32073028);宁波市自然科学基金(2022J165)

Depression Detection Based on Cross User Audio Domain Adaptation Network

WU Wei¹, MA Longhua¹^,², ZHAO Xianghong²()

1. School of Information Science and Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China
2. School of Information Science and Engineering,Ningbo Tech University,Ningbo 315100,China

Received:2023-06-25 Revised:2023-07-16 Online:2025-01-15 Published:2025-01-06
Supported by:
National Natural Science Foundation of China(61972350);National Natural Science Foundation of China(32073028);Ningbo Natural Science Foundation(2022J165)

摘要/Abstract

摘要：

由于抑郁症的检测方式主观性较强,因此使用用户语音诊断抑郁症已成为一种较具有潜力的辅助方式,但不同用户的语音信号存在差异。文中提出了一个跨用户语音域适应网络(Cross User Audio Domain Adaptation Network, CUADAN)来检测抑郁症。从语音中提取可视化的梅尔频谱,利用CUADAN模型的特征提取器从梅尔频谱中获取更深层次的抑郁特征。由于源域和目标域中包含不同健康用户和抑郁用户的语音特征,因此利用CUADAN模型的域分类器在不同用户数据之间进行域适应,从而通过已有分类器对未知用户进行检测。实验结果表明,CUADAN模型的抑郁症检测准确率更高,其平均准确率达到81.0±2.4%。因此,CUADAN模型可以有效削弱不同用户语音之间的差异性,提高跨用户抑郁症检测的准确率。

关键词: 域适应, 抑郁症检测, CUADAN, 语音, 跨用户, 梅尔频谱, 特征提取, 削弱差异性

Abstract:

Because of the subjective detection of depression, the use of user voice diagnosis of depression has become a more potential auxiliary way. However, the speech signals of different users are different. In this study, a CUADAN(Cross User Audio Domain Adaptation Network) is proposed to detect depression. Visual Mel spectrograms are extracted from the audio, and the feature extractor of the CUADAN model is used to extract deeper depression features from the Mel spectrograms. Since the source domain and target domain contain the voice features of different healthy users and depressed users, the domain classifier of CUADAN model is used to perform domain adaptation between different user data, so that unknown users can be detected by existing classifiers. The experimental results show that the CUADAN model has a higher depression detection accuracy, with an average accuracy of 81.0±2.4%. Therefore, the CUADAN model can effectively weaken the differences between different users' voices and improve the accuracy of cross-user depression detection.

Key words: domain adaptation, depression detection, CUADAN, audio, cross-user, Mel spectrogram, feature extraction, weakening differences

中图分类号:

TP181

吴伟, 马龙华, 赵祥红. 基于跨用户语音域适应网络的抑郁症检测[J]. 电子科技, 2025, 38(1): 88-94.

WU Wei, MA Longhua, ZHAO Xianghong. Depression Detection Based on Cross User Audio Domain Adaptation Network[J]. Electronic Science and Technology, 2025, 38(1): 88-94.

图/表 7

表1

图1

图2

图3

图4

表2

图5

参考文献 20

[1]	Fried E I, Flake J K, Robinaugh D J. Revisiting the theoretical and methodological foundations of depression measurement[J]. Nature Reviews Psychology, 2022, 1(6):358-368. doi: 10.1038/s44159-022-00050-2 pmid: 38107751
[2]	陈方方, 吴明飞, 张佳俐. 认知情绪调节策略对青少年单相抑郁和双相抑郁的影响[J]. 中华行为医学与脑科学杂志, 2022, 31(12):1086-1091.
	Chen Fangfang, Wu Mingfei, Zhang Jiali. The impact of cognitive emotion regulation strategies on unipolar and bipolar depression in adolescents[J]. Chinese Journal of Behavioral Medicine and Brain Science, 2022, 31(12):1086-1091.
[3]	Dubovsky S L, Ghosh B M, Serotte J C, et al. Psychotic depression:Diagnosis,differential diagnosis and treatment[J]. Psychotherapy and Psychosomatics, 2021, 90(3):160-177. doi: 10.1159/000511348 pmid: 33166960
[4]	Carrozzino D, Patierno C, Fava G A, et al. The hamilton rating scales for depression:A critical review of clinimetric properties of different versions[J]. Psychotherapy and Psychosomatics, 2020, 89(3):133-150. doi: 10.1159/000506879 pmid: 32289809
[5]	Ubukata S, Ueda K, Fujimoto G, et al. Extracting apathy from depression syndrome in traumatic brain injury by using a clustering method[J]. The Journal of Neuropsychiatry and Clinical Neurosciences, 2022, 34(2):158-167.
[6]	Stasak B, Joachim D, Epps J. Breaking age barriers with automatic voice-based depression detection[J]. IEEE Pervasive Computing, 2022, 21(2):10-19.
[7]	Haick H, Tang N. Artificial intelligence in medical sensors for clinical decisions[J]. ACS Nano, 2021, 15(3):3557-3567. doi: 10.1021/acsnano.1c00085 pmid: 33620208
[8]	Huang Z, Epps J, Joachim D, et al. Natural language processing methods for acoustic and landmark event-based features in speech-based depression detection[J]. IEEE Journal of Selected Topics in Signal Processing, 2019, 14(2):435-448.
[9]	Calić G, Petrović-Lazić M, Mentus T, et al. Acoustic features of voice in adults suffering from depressio[J]. Psihološka Istraživanja, 2022, 25(2):183-203.
[10]	Ding Y, Zhong S, Hua L. Automatic recognition of student emotions based on deep neural network and its application in depression detection[J]. Journal of Medical Imaging and Health Informatics, 2020, 10(11):2634-2641.
[11]	Zhuang F, Qi Z, Duan K, et al. A comprehensive survey on transfer learning[J]. Proceedings of the IEEE, 2020, 109(1):43-76.
[12]	Long M, Cao Y, Wang J, et al. Learning transferable features with deep adaptation networks[C]. Lille: International Conference on Machine Learning, 2015:97-105.
[13]	Cai H, Yuan Z, Gao Y, et al. A multi-modal open dataset for mental-disorder analysis[J]. Scientific Data, 2022, 9(1):178-178. doi: 10.1038/s41597-022-01211-x pmid: 35440583
[14]	Nguyen M T, Lin W W, Huang J H. Heart sound classification using deep learning techniques based on Log-Mel spectrogram[J]. Circuits,Systems and Signal Processing, 2023, 42(1):344-360.
[15]	Park J, Moon N. Design and implementation of attention depression detection model based on multimodal analysis[J]. Sustainability, 2022, 14(6):1-15.
[16]	詹雁, 张娟. 一种结构感知损失的域适应深度估计方法[J]. 电子科技, 2020, 33(12):12-16.
	Zhan Yan, Zhang Juan. A domain adaptive depth estimation method for structural perception loss[J]. Electronic Science and Technology, 2020, 33(12):12-16.
[17]	Finjan R H, Rasheed A S, Hashim A A, et al. Arabic handwritten digits recognition based on convolutional neural networks with ResNet-34 model[J]. Indonesian Journal of Electrical Engineering and Computer Science, 2021, 21(1):174-178.
[18]	林潮威, 李菲菲, 陈虬. 基于深度卷积特征的场景全局与局部表示方法[J]. 电子科技, 2022, 35(4):20-27.
	Lin Chaowei, Li Feifei, Chen Qiu. Globaland local scene representation method based on deep convolutional features[J]. Electronic Science and Technology, 2022, 35(4):20-27.
[19]	Ganin Y, Ustinova E, Ajakan H, et al. Domain-adversarial training of neural networks[J]. The Journal of Machine Learning Research, 2016, 17(1):2096-2030.
[20]	Kobak D, Linderman G C. Initialization is critical for preserving global data structure in both t-SNE and UMAP[J]. Nature Biotechnology, 2021, 39(2):156-157.

特征	抑郁组	健康组
人数(男∶女)	16∶7	20∶9
年龄	16~56	18~55
接受教育年数	6~19	12~19

模型	第1组 /%	第2组 /%	第3组 /%	第4组 /%	平均准确率及标准差/%
SVM	67.2	57.5	48.9	65.8	59.8±7.3
LR	63.2	44.8	57.4	60.7	56.5±7.1
DenseNet	71.6	63.5	68.4	70.8	68.6±3.2
ConvLSTM	74.7	67.8	69.2	71.7	70.9±2.6
CUADAN	84.4	78.9	78.6	82.2	81.0±2.4

基于跨用户语音域适应网络的抑郁症检测

Depression Detection Based on Cross User Audio Domain Adaptation Network

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 20

相关文章 15

Metrics

本文评价

推荐阅读 0

[1]	王显云, 窦姗姗, 程楚皓. 基于对数处理机制和时频掩蔽估计的语音增强[J]. 电子科技, 2025, 38(1): 45-51.
[2]	韩瀛昊, 李菲菲. 基于关键区域特征提取与二阶段分类网络的场景识别方法[J]. 电子科技, 2024, 37(7): 25-32.
[3]	李静, 彭涛. 基于基音延迟隐写的移动语音信道数据传输[J]. 电子科技, 2024, 37(7): 66-71.
[4]	王圣雄, 刘瑞安, 燕达. 基于改进生成对抗网络的图像风格迁移算法[J]. 电子科技, 2024, 37(6): 36-43.
[5]	庞江飞, 孙占全. 基于Transformer的多编码器端到端语音识别[J]. 电子科技, 2024, 37(4): 1-7.
[6]	孙红,张玉香. 基于多特征门控反馈残差网络的超分辨率图像重建算法[J]. 电子科技, 2023, 36(4): 65-70.
[7]	高于科,章伟,胡陟,江鹏伟. 基于RFNA和改进LBD的镜像线特征匹配方法[J]. 电子科技, 2023, 36(10): 32-38.
[8]	张漫秸,杨芳艳,季云峰. 球类运动中人体姿态估计研究进展[J]. 电子科技, 2023, 36(1): 28-37.
[9]	张乔木,钟倩文,孙明,罗文成,柴晓冬. 复杂环境下弓网接触位置动态监测方法研究[J]. 电子科技, 2022, 35(8): 66-72.
[10]	孙抗,轩旭阳,刘鹏辉,赵来军,龙洁. 小样本下基于CNN-DCGAN的电缆局部放电模式识别方法[J]. 电子科技, 2022, 35(7): 7-13.
[11]	李辉,景浩,严康华,徐良浩. 基于卷积循环网络与非局部模块的语音增强方法[J]. 电子科技, 2022, 35(3): 8-15.
[12]	伍伟嘉,杨俭,袁天辰,邵志慧. 基于支持向量机的轨道结构病害识别研究[J]. 电子科技, 2022, 35(2): 27-33.
[13]	李辉,王一丞. 基于CNNCIFG-Attention模型的文本情感分类[J]. 电子科技, 2022, 35(2): 46-51.
[14]	邵志慧,杨俭,袁天辰,伍伟嘉. 基于排列熵和支持向量机的轨枕病害诊断[J]. 电子科技, 2022, 35(2): 52-58.
[15]	苏波,柴自强,王莉,崔帅华. 基于姿态估计的八段锦序列动作识别与评估[J]. 电子科技, 2022, 35(12): 84-90.