Discriminating Between Imagined Speech and Non-Speech Tasks Using EEG.

People who are severely disabled (e.g Locked-in patients) need a communication tool translating their thoughts using their brain signals. This technology should be intuitive and easy to use. To this line, this study investigates the possibility of discriminating between imagined speech and two types of non-speech tasks related to either a visual stimulus or relaxation. In comparison to previous studies, this work examines a variety of different words with only single imagination in each trial. Moreover, EEG data are recorded from a small number of electrodes using a low-cost portable EEG device. Thus, our experiment is closer to what we want to achieve in the future as communication tool for locked-in patients. However, this design makes the EEG classification more challenging due to a higher level of noise and variations in EEG signals. Spectral and temporal features, with and without common spatial filtering, were used for classifying every imagined word (and for a group of words) against the non-speech tasks. The results show the potential for discriminating between each imagined word and non-speech tasks. Importantly, the results are different between subjects using different features showing the need for having subject specific features.