Arousal and Valence Recognition of Affective Sounds Based on Electrodermal Activity

Physiological sensors and interfaces for mental healthcare are becoming of great interest in research and commercial fields. Specifically, biomedical sensors and related ad hoc signal processing methods can be profitably used for supporting objective, psychological assessments. However, a simple system able to automatically classify the emotional state of a healthy subject is still missing. To overcome this important limitation, we here propose the use of convex optimization-based electrodermal activity (EDA) framework and clustering algorithms to automatically discern arousal and valence levels induced by affective sound stimuli. EDA recordings were gathered from 25 healthy volunteers, using only one EDA sensor to be placed on fingers. Standardized stimuli were chosen from the International Affective Digitized Sound System database, and grouped into four different levels of arousal (i.e., the levels of emotional intensity) and two levels of valence (i.e., how unpleasant/pleasant a sound can be perceived). Experimental results demonstrated that our system is able to achieve a recognition accuracy of 77.33% on the arousal dimension, and 84% on the valence dimension.