Overt and Covert Speech-State Classification from Tripolar Concentric Ring Electrode EEG: A 4 mm and 10 mm...

Electroencephalography (EEG) speech-state classification is a noninvasive method to identify the brain's overt and covert behavior. The EEG achieved through the scalp is contaminated by artifacts and volume conduction and suffers from poor spatial resolution. We present the first application of small 4 mm tripolar concentric ring electrodes (TCREs) to record the overt and covert EEG signals. In addition, EEG signals were recorded for overt and covert behavior using a 10 mm TCRE as a reference. Data were recorded from 16 participants (4 mm, n=8; 10 mm, n=8). Each participant completed 200 picture-naming trials (100 overt and 100 covert). The analysis includes task-to-baseline root mean square ratios, power spectral density, time frequency representation analysis and participant-wise overt and covert classification using linear discriminant analysis, support vector machines, and EEGNet. The overt speech produced higher task-to-baseline ratios than covert speech. In addition, EEGNet achieved the highest mean accuracy (84.06±15.57%), followed by SVM (81.56±14.23%) and LDA (76.56±15.30%).