Affective computation on EEG correlates of emotion from musical and vocal stimuli

Affective interface that acquires and detects the emotion of the user can potentially enhance the human-computer interface experience. In this paper, an affective brain-computer interface (ABCI) is proposed to perform affective computation on electroencephalogram (EEG) correlates of emotion. The proposed ABCI extracts EEG features from subjects while exposed to 6 emotionally-related musical and vocal stimuli using kernel smoothing density estimation (KSDE) and Gaussian mixture model probability estimation (GMM). A classification algorithm is subsequently used to learn and classify the extracted EEG features. An inter-subject validation study is performed on healthy subjects to assess the performance of ABCI using a selection of classification algorithms. The results show that ABCI that employed the Bayesian network and the one-rule classifier yielded a promising inter-subject validation accuracy of 90%.