Advancing Mental Health Diagnostics: A Novel Deep Learning Approach to Vocal Biomarker Identification for Stress Detection in a Korean Population (Preprint)

BACKGROUND Escalating mental health concerns exacerbated by the coronavirus disease (COVID-19) and rapid societal shifts have made efficient monitoring of mental stress crucial. Chronic mental stress affects physical and psychological health, necessitating timely identification and intervention OBJECTIVE To explore the viability of vocal biomarkers in detecting stress levels among healthy Korean employees and to contribute to digital healthcare solutions. METHODS A multicenter clinical trial was conducted by collecting voice recordings of 113 healthy Korean employees under relaxed and stress-induced conditions. A deep learning architecture used for speaker identification was employed to develop stress prediction scores RESULTS The proposed model demonstrated an accuracy of 70% for stress detection. This highlights the potential of vocal biomarkers in digital healthcare to offer a convenient and effective means for individuals to self-monitor and manage their stress levels. CONCLUSIONS These findings underscore the potential of voice-based mental stress assessment among Koreans, emphasizing the importance of research on vocal biomarkers across linguistic demographics