Abstract
BACKGROUND: Digital tools are increasingly widespread in healthcare, particularly in the fields of emotion recognition and mental health assessment. OBJECTIVES: This study evaluated whether an artificial intelligence (AI) voice emotion recognition (VER) app could identify nurses' emotions and explored its associations with their background and health conditions. METHODS: The emotions of 349 clinical nurses at a medical center in northern Taiwan were analyzed using an AI VER app and several standardized psychological questionnaires. To control for potential confounding variables, demographic and health-related factors including age, gender, work experience, exercise habits, and history of physical symptoms were collected and statistically adjusted in correlation analyses. Convergent validity was tested with Pearson's correlations, and test-retest reliability was evaluated in 30 nurses using intraclass correlation coefficients (ICCs). RESULTS: Significant correlations were observed between app-derived emotions and standard scales (anger: Novaco Anger Inventory-Short Form, r = .42; fear: Perceived Stress Scale, r = .41; happiness: Oxford Happiness Questionnaire, r = .45; and sadness: Beck Depression Inventory-II, r = .47; all p < .001). Multiple regression identified significant lifestyle and health predictors of emotions: less exercise predicted higher anger (β = .11, p = .025), peptic ulcers predicted greater fear (β = .19, p < .001), daily coffee predicted higher happiness (β = .11, p = .041), and irregular menstrual cycles predicted lower happiness (β = -.13, p = .014) and greater sadness (β = .30, p < .001). The AI VER app demonstrated good test-retest reliability (ICC = 0.73-0.80). CONCLUSION: Peptic ulcers, irregular menstrual cycles, and lack of exercise were associated with negative emotions such as fear, sadness, and anger. The AI VER app could objectively detect these emotional patterns in nurses, helping to identify emotional fluctuations early and support timely mental healthcare.