Classifying siren-sound mental rehearsal and covert production vs. idle state towards onset detection in brain-computer interfaces

This research investigated the potential of a new method for onset detection towards asynchronous BCIs. Siren sound covert production and recall were classified against the idle (no task) state in an off-line system. Wavelet packet decomposition was employed for feature extraction and a Support Vector Machine (SVM) was used for classification. Three window segments lengths were tested (1s, 2s and 3s), but a Wilcoxon test showed that there is no significant difference between the results for different segment lengths. Using 1s window length, the system achieved 76.88%, 79.58%, 76.67%, 80.2% and 82.71% true positive accuracy for five subjects, respectively.