Meaning based covert speech classification for brain-computer interface based on electroencephalography

In this study, we investigated that whether covertly spoken words with different meaning are discriminable during electroencephalography (EEG) recording. Neural activities were recorded from 30 channel 10-20 system electrodes. By employing a paired T-test, we briefly identify the difference in spatio-spectro-temporal characteristics between two categories of meaning (number and face). EEG features were then classified by support vector machine. On average, 71.69% of the trials were correctly classified. After extract optimized features using support vector machine based recursive feature elimination, the accuracy was improved up to 92.46%. Our preliminary results shed light on the construction of meaning based speech brain-computer interface.