Development of an objective early detection model for depressive symptoms using voice emotion analysis technology:
Naomichi Tani1, Yoshihiro Takao2, Sakihito Noro2
1Department of Ergonomics, Institute of Industrial Ecological Sciences, University of Occupational and Environmental Health, Japan, 1-1 Iseigaoka, Yahatanishi-ku, Fukuoka, Kitakyushu 807-8555, Japan.
Objectives:
Voice and emotional analyses have gained attention in the diagnosis and monitoring of depression in clinical settings. However, evidence supporting its use for early detection in occupational health is lacking. This study aimed to develop a predictive model to identify early depressive symptoms in workers using voice and emotional analyses.
Methods:
A prospective cohort study was conducted with 62 call center workers in Kumamoto Prefecture, Japan. The participants' voices were automatically recorded during routine operations and analyzed using a voice and emotional analysis system based on Layered Voice Analysis. Depressive symptoms were assessed at 4 time points over 12 weeks using the Center for Epidemiologic Studies Depression Scale. Recursive Feature Elimination identified optimal voice features, while logistic regression was used to calculate the probability scores and build a predictive model for depressive symptoms. Predictive accuracy was evaluated using receiver operating characteristic curves and the area under the curve.
Results:
The predictive model's accuracy reached 0.783 (95% CI, 0.691-0.875) for the area under the curve, with a sensitivity of 0.649, a 1 - specificity of 0.174, and a cutoff value of 0.334. Individuals with composite voice indicators above the determined cutoff were significantly more likely to exhibit depressive symptoms 1 month later (odds ratio = 7.78; 95% CI, 3.27-18.5).
Conclusions:
This study suggests that voice and emotional analysis can serve as an objective tool for the early identification of depressive symptoms in workplace settings. Accumulating real-world evidence from observational studies in diverse occupational populations is required to support broader implementation.
