查看文章

Human’s emotional states affect their utterances which are generated through vocal cord vibrations. Accurate recognition of these emotional states encoded in human speech signals is critical and can be leveraged for mental health purposes. such as assisting practitioners in their assessments and decisionmaking, improving therapy effectiveness, safety monitoring of patients, and clinical training. Although there are existing works on speech emotion recognition, very few works address speech emotion recognition from a mental health perspective. This paper presents the results of our preliminary analysis that demonstrate the feasibility of automatic speech emotion recognition for mental health purposes. We used five machine learning paradigms for classifying emotions and evaluated their performance by focusing on their effectiveness in capturing human emotions using custom and benchmark databases, including TESS, EMO-DB, and RAVDESS. SVM demonstrated superior performance in overlapping settings based on F1-value and achieved 74% accuracy in RAVDESS and the custom datasets. We believe this research could be the initial step towards a fully implemented intelligent support service for mental health.