Construction of recognition models for subthreshold depression based on multiple machine learning algorithms and vocal emotional characteristics

doi:10.12122/j.issn.1673-4254.2025.04.05

Abstract

Abstract:

Objective To construct vocal recognition classification models using 6 machine learning algorithms and vocal emotional characteristics of individuals with subthreshold depression to facilitate early identification of subthreshold depression. Methods We collected voice data from both normal individuals and participants with subthreshold depression by asking them to read specifically chosen words and texts. From each voice sample, 384-dimensional vocal emotional feature variables were extracted, including energy feature, Meir frequency cepstrum coefficient, zero cross rate feature, sound probability feature, fundamental frequency feature, difference feature. The Recursive Feature Elimination (RFE) method was employed to select voice feature variables. Classification models were then built using the machine learning algorithms Adaptive Boosting (AdaBoost), Random Forest (RF), Linear Discriminant Analysis (LDA), Logistic Regression (LR), Lasso Regression (LRLasso), and Support Vector Machine (SVM), and the performance of these models was evaluated. To assess generalization capability of the models, we used real-world speech data to evaluate the best speech recognition classification model. Results The AdaBoost, RF, and LDA models achieved high prediction accuracies of 100%, 100%, and 93.3% on word-reading speech test set, respectively. In the text-reading speech test set, the accuracies of the AdaBoost, RF, and LDA models were 90%, 80%, and 90%, respectively, while the accuracies of the other 3 models were all below 80%. On real-world word-reading and text-reading speech data, the classification models using AdaBoost and Random Forest still achieved high predictive accuracies (91.7% and 80.6% for AdaBoost and 86.1% and 77.8% for Random, respectively). Conclusion Analyzing vocal emotional characteristics allows effective identification of individuals with subthreshold depression. The AdaBoost and RF models show excellent performance for classifying subthreshold depression individuals, and may thus potentially offer valuable assistance in the clinical and research settings.

Key words: subthreshold depression recognition, phonological and emotional characteristics, machine learning, AdaBoost, random forest

Meimei CHEN, Yang WANG, Huangwei LEI, Fei ZHANG, Ruina HUANG, Zhaoyang YANG. Construction of recognition models for subthreshold depression based on multiple machine learning algorithms and vocal emotional characteristics[J]. Journal of Southern Medical University, 2025, 45(4): 711-717.

Figures/Tables 6

References 31

1	关茜, 周小芳, 李福凤. 闻诊中医理论基础及现代化研究进展[J]. 中华中医药杂志, 2022, 37(4): 2134-6.
2	Yamamoto M, Takamiya A, Sawada K, et al. Using speech recognition technology to investigate the association between timing-related speech features and depression severity[J]. PLoS One, 2020, 15(9): e0238726.
3	Liang L, Wang Y, Ma H, et al. Enhanced classification and severity prediction of major depressive disorder using acoustic features and machine learning[J]. Front Psychiatry, 2024, 15: 1422020.
4	Zhao Q, Fan HZ, Li YL, et al. Vocal acoustic features as potential biomarkers for identifying/diagnosing depression: a cross-sectional study[J]. Front Psychiatry, 2022, 13: 815678.
5	König A, Tröger J, Mallick E, et al. Detecting subtle signs of depression with automated speech analysis in a non-clinical sample[J]. BMC Psychiatry, 2022, 22(1): 830.
6	Koops S, Brederoo SG, de Boer JN, et al. Speech as a biomarker for depression[J]. CNS Neurol Disord Drug Targets, 2023, 22(2): 152-60.
7	Little B, Alshabrawy O, Stow D, et al. Deep learning-based automated speech detection as a marker of social functioning in late-life depression[J]. Psychol Med, 2021, 51(9): 1441-50.
8	袁钦湄, 王星, 帅建伟, 等. 基于人工智能技术的抑郁症研究进展[J]. 中国临床心理学杂志, 2020, 28(1): 82-6. DOI: 10.16128/j.cnki.1005-3611.2020.01.019
9	罗德虎, 冉启武, 杨超, 等. 语音情感识别研究综述[J]. 计算机工程与应用, 2022, 58(21): 40-52. DOI: 10.3778/j.issn.1002-8331.2206-0352
10	Judd LL, Rapaport MH, Paulus MP, et al. Subsyndromal symptomatic depression: a new mood disorder[J]? J Clin Psychiatry, 1994, 55(): 18-28.
11	沈渔邨. 精神病学[M]. 北京: 北京医科大学出版社, 2002.
12	王一牛, 周立明, 罗跃嘉. 汉语情感词系统的初步编制及评定[J]. 中国心理卫生杂志, 2008, 22(8): 608-12.
13	Eyben F, Wöllmer M, Schuller B. Opensmile: the Munich versatile and fast open-source audio feature extractor[C]//Proceedings of the 18th ACM International Conference on Multimedia. Firenze Italy. ACM, 2010: 1459-1462.
14	Jeon H, Oh S. Hybrid-recursive feature elimination for efficient feature selection[J]. Appl Sci, 2020, 10(9): 3211.
15	Cervantes J, Garcia-Lamont F, Rodríguez-Mazahua L, et al. A comprehensive survey on support vector machine classification: Applications, challenges and trends[J]. Neurocomputing, 2020, 408: 189-215.
16	向进勇, 王振华, 邓芸芸. 基于随机森林算法的机器学习分类研究综述[J]. 人工智能与机器人研究, 2024, 13(1): 143-52.
17	奚丽婧, 郭昭艳, 杨雪珂, 等. LASSO及其拓展方法在回归分析变量筛选中的应用[J]. 中华预防医学杂志, 2023, 57(1): 107-11.
18	Song X, Liu XY, Liu F, et al. Comparison of machine learning and logistic regression models in predicting acute kidney injury: a systematic review and meta-analysis[J]. Int J Med Inform, 2021, 151: 104484.
19	Zhao SP, Zhang B, Yang J, et al. Linear discriminant analysis[J]. Nat Rev Meth Primers, 2024, 4: 70.
20	Taherkhani A, Cosma G, McGinnity TM. AdaBoost-CNN: an adaptive boosting algorithm for convolutional neural networks to classify multi-class imbalanced datasets using transfer learning[J]. Neurocomputing, 2020, 404: 351-66.
21	Rainio O, Teuho J, Klén R. Evaluation metrics and statistical tests for machine learning[J]. Sci Rep, 2024, 14: 6086.
22	Menne F, Dörr F, Schräder J, et al. The voice of depression: speech features as biomarkers for major depressive disorder[J]. BMC Psychiatry, 2024, 24(1): 794.
23	Di Y, Wang J, Liu X, et al. Combining polygenic risk score and voice features to detect major depressive disorders[J]. Front Genet, 2021, 12: 761141.
24	Di Y, Wang J, Li W, et al. Using i-vectors from voice features to identify major depressive disorder[J]. J Affect Disord, 2021, 288: 161-6.
25	吴朝毅, 王振. 抑郁症情绪失调的动态特征：情绪动力学的视角[J]. 心理科学进展, 2024, 32(2): 364-385.
26	刘振焘, 向春妮, 刘陈陵, 等. 基于语音的抑郁检测研究综述[J]. 信号处理, 2023, 39(4): 616-31.
27	Rejaibi E, Komaty A, Meriaudeau F, et al. MFCC-based Recurrent Neural Network for automatic clinical depression recognition and assessment from speech[J]. Biomed Signal Process Contr, 2022, 71: 103107.
28	Yokota K, Ishikawa S, Takezaki K, et al. Numerical analysis and physical consideration of vocal fold vibration by modal analysis[J]. J Sound Vib, 2021, 514: 116442.
29	Ozdas A, Shiavi RG, Silverman SE, et al. Investigation of vocal jitter and glottal flow spectrum as possible cues for depression and near-term suicidal risk[J]. IEEE Trans Biomed Eng, 2004, 51(9): 1530-40.
30	Wang Y, Liang L, Zhang Z, et al. Fast and accurate assessment of depression based on voice acoustic features: a cross-sectional and longitudinal study[J]. Front Psychiatry, 2023, 14: 1195276.
31	Hammoud M, Getahun MN, Baldycheva A, et al. Machine learning-based infant crying interpretation[J]. Front Artif Intell, 2024, 7: 1337356.

Group	Case	Age (year)	CES-D (min)	HAM-D (min)
Health	60	18-30	0-14	0-6
Subthreshold depression	50	18-30	16-19	7-17

Group	Case	Age (year)	CES-D (min)	HAM-D (min)
Health	60	18-30	0-14	0-6
Subthreshold depression	50	18-30	16-19	7-17

Speech feature variable	Definition	Speech feature type	Variable selection order
pcm_RMSenergy_sma_amean	Mean of root mean square energy	Energy feature	1
pcm_RMSenergy_sma_linregc2	Linear regression coefficient of Root Mean Square Energy, describing the long-term trend of energy	Energy feature	2
pcm_RMSenergy_sma_stddev	Standard deviation of Root Mean Square Energy, reflecting the degree of energy variation	Energy feature	3
pcm_zcr_sma_linregc1	Linear regression coefficient of zero-crossing rate, reflecting the rate of change in the speech signal	Zero cross rate feature	4
pcm_zcr_sma_linregerrQ	Error of zero-crossing rate linear regression, reflecting the stability of signal changes	Zero cross rate feature	5
voiceProb_sma_linregc1	Linear regression coefficient of speech probability, especially regarding changes in syllables and stress	Sound source feature	6
pcm_fftMag_mfcc_sma_de^[1]_amean	Mean of the first dimension Mel-Frequency Cepstral Coefficients (MFCC), providing an overall description of the signal's spectral characteristics.	Spectral feature	7
pcm_fftMag_mfcc_sma_de^[1]_linregc2	Linear regression coefficient 2 of the first dimension MFCC, reflecting the trend of spectral feature changes.	Spectral feature	8
pcm_zcr_sma_de_skewness	Skewness of zero-crossing rate, reflecting the transient characteristics of the signal.	Zero cross rate feature	9
pcm_zcr_sma_de_kurtosis	Kurtosis of zero-crossing rate, providing information about the sharpness of the signal.	Zero cross rate feature	10

Speech feature variable	Definition	Speech feature type	Variable selection order
pcm_RMSenergy_sma_amean	Mean of root mean square energy	Energy feature	1
pcm_RMSenergy_sma_linregc2	Linear regression coefficient of Root Mean Square Energy, describing the long-term trend of energy	Energy feature	2
pcm_RMSenergy_sma_stddev	Standard deviation of Root Mean Square Energy, reflecting the degree of energy variation	Energy feature	3
pcm_zcr_sma_linregc1	Linear regression coefficient of zero-crossing rate, reflecting the rate of change in the speech signal	Zero cross rate feature	4
pcm_zcr_sma_linregerrQ	Error of zero-crossing rate linear regression, reflecting the stability of signal changes	Zero cross rate feature	5
voiceProb_sma_linregc1	Linear regression coefficient of speech probability, especially regarding changes in syllables and stress	Sound source feature	6
pcm_fftMag_mfcc_sma_de^[1]_amean	Mean of the first dimension Mel-Frequency Cepstral Coefficients (MFCC), providing an overall description of the signal's spectral characteristics.	Spectral feature	7
pcm_fftMag_mfcc_sma_de^[1]_linregc2	Linear regression coefficient 2 of the first dimension MFCC, reflecting the trend of spectral feature changes.	Spectral feature	8
pcm_zcr_sma_de_skewness	Skewness of zero-crossing rate, reflecting the transient characteristics of the signal.	Zero cross rate feature	9
pcm_zcr_sma_de_kurtosis	Kurtosis of zero-crossing rate, providing information about the sharpness of the signal.	Zero cross rate feature	10

Model algorithm	Training set				Test set
Model algorithm	Accuracy	Sensitivity	Specificity	ROC AUC	Accuracy	Sensitivity	Specificity	ROC AUC
AdaBoost	100%	100%	100%	100%	100%	100%	100%	100%
RF	100%	100%	100%	100%	100%	100%	100%	100%
LDA	95.2%	96.7%	93.8%	97.8%	93.3%	85.7%	100%	87.5%
LRLasso	82.3%	96.7%	68.8%	79.5%	93.3%	100%	87.5%	98.2%
SVM	82.3%	96.7%	68.8%	75.8%	86.7%	100%	75%	91.1%
LR	82.3%	96.7%	68.8%	75.7%	86.7%	100%	75%	85.7%