Classifying speech related vs. idle state towards onset detection in brain-computer interfaces overt, inhibited overt, and covert speech sound production vs. idle state

Onset detection is one of the main issues towards self-paced BCIs that can be used outside research settings. For this reason, this paper suggests a potential solution for onset detection problem by discriminating between speech related events. In this study, overt, inhibited overt and covert states were tested to classify from idle state in an off-line setting. Autoregressive model coefficients were used for feature extraction. The results showed that covert speech (vs. idle state) performed the best for all 4 participants. The true positive accuracies were 82.41%, 81.20%, 85.12% and 74.72%, respectively. The bit-transfer rates were 32.95, 16.24, 34.05 and 22.42 per minute, respectively. Compared to a previous study [1], which achieved around 73% accuracy with motor imagery versus idle, this study gave us satisfactory results.