A brain-computer interface system for online spelling

We consider the problem of spelling through an electroencephalography (EEG) based brain-computer interface and present a complete system with associated algorithms for automatic online classification as well as offline classification. In our system we use a flexible visual stimulus mechanism adaptable to user preferences that we have designed. This mechanism aims to exploit the P300 wave in the EEG signal, generated in response to unpredictable stimuli. Through training sessions, we learn the EEG patterns of the subjects in the presence and absence of the P300 wave in the context of a spelling experiment. We present EEG signal processing and classification algorithms for online automated decision making on the character targeted by the subject. We use a classifier based on Bayes linear discriminant analysis (BLDA) and propose a greedy approach for increasing the spelling rate. We have run numerous offline and online experiments demonstrating the effectiveness of our system and performance improvements it provides over results published in the literature.