Towards an Artificial Therapy Assistant - Measuring Excessive Stress from Speech

The measurement of (excessive) stress is still a challenging endeavor. Most tools rely on either introspection or expert opinion and are, therefore, often less reliable or a burden on the patient. An objective method could relieve these problems and, consequently, assist diagnostics. Speech was considered an excellent candidate for an objective, unobtrusive measure of emotion. True stress was successfully induced, using two storytelling sessions performed by 25 patients suffering from a stress disorder. When reading either a happy or a sad story, different stress levels were reported using the Subjective Unit of Distress (SUD). A linear regression model consisting of the high-frequency energy, pitch, and zero crossings of the speech signal was able to explain 70% of the variance in the subjectively reported stress. The results demonstrate the feasibility of an objective measurement of stress in speech. As such, the foundation for an Artificial Therapeutic Agent is laid, capable of assisting therapists through an objective measurement of experienced stress.