Emotion recognition from sound stimuli based on back-propagation neural networks and electroencephalograms.

This research aims to explore the feasibility of using back-propagation (BP) neural networks and electroencephalograms (EEGs) to recognize the emotional reactions induced by sound stimuli in the dimensions of pleasure and arousal, as well as compare the recognition performance of each method on these two dimensions. It could provide an aided design on choosing proper sounds to induce or regulate individuals' emotional states under specific situations for potential users at the design stage. Emotional reactions to different sound stimuli are investigated by Self-Assessment Manikin. The results of BP neural network indicate that the arousal predictions are more satisfactory than the pleasure predictions, and the recognition rates can be improved by optimizing input parameters. EEG signals induced by sound stimuli are recorded. The results show that when induced by each pleasant sound, the Average Power of Electroencephalogram of the α wave in the left frontal pole electrode is significantly lower than that in the right frontal pole electrode, while when induced by each unpleasant sound, the former is significantly higher than the latter. This finding indicates that pleasant and unpleasant sounds can be identified based on the asymmetry of the α wave between the left and right frontal pole electrodes.