Several current brain–computer interface (BCI) systems are based on imagined speech. This means that these systems are controlled only by thinking about a speech without verbally expressing it. Imagined speech recognition using electroencephalogram (EEG) signals is much more convenient than other methods such as electrocorticogram (ECoG), due to its easy, non-invasive recording. So, we proposed anapproach for EEG classification of imagined speech with high accuracy and efficiency. In this work, correntropy spectral density (CSD) matrices are evaluated for EEG signals obtained from different channels,and the distances between these matrices are considered as measures for imagined speech recognition. Riemannian distance benefits from simplicity and accuracy and it has achieved high scores in BCI com-petitions. Also, in this work, channel selection and frequency band detection during imagined speech is evaluated with statistical methods. The “Kara One” database is used in this research that includes EEG signals of eight subjects during imagined speech of four English words. We evaluated the proposed approach in comparison with the results of other imagined speech classification methods. Average classification accuracy of the proposed method is 90.25% during imagined speech for all subjects in KARA One database.The simulation results of this paper show the efficiency and accuracy of Riemannian distance of CSD and the superiority of the proposed method over other methods for imagined speech classification.

Keywords