Real-Time Silent Speech BCI with Japanese Hiragana

摘要

Brain-Computer interfaces (BCIs) are often seen as a way to bypass limbs and might be used by people physically unable to speak. BCI have been proven accurate enough to be useful but they are used with a heavy montage and long computational time. By focusing only on one electrode, placed on the Broca's area, and by using multiple short computational-time methods like extreme learning machine (ELM) classifiers, we tried to have the best silent vowel recognition accuracy in the shortest time. We managed to get over 50% accuracy within 3 seconds.