Recognition of Imagined Speech using Electroencephalogram Signals

摘要

Mind reading, or imagined speech recognition, has many real-world applications including silent communicationbetween soldiers in a combat zone, lie-detectors, and even communication with stroke victims who are fullycoherent but unable to produce verbal speech. However, while previous research has shown that using electroen-cephalography (EEG) to read imagined speech is certainly possible, it is currently impractical because each EEGdevice can be calibrated for only one person. No calibration test has been developed to allow the same EEGdevice to accurately read the imagined speech of di erent people. A calibration test is necessary because di erentpeople have di erent neurons ring even when thinking of the same letter, word, syllable, etc. For example, theneurons that re when person A thinks of the letter 'a' is di erent than those red when person B thinks of thesame letter. In this paper, and as a rst step we demonstrate the possibility of reading a brain using noninvasive(EEG) technology. Further an algorithm is developed to calibrate the same device for di erent subjects. Thisis accomplished rst by gathering raw EEG data, performing pre-processing on that data, performing featureextraction, and nally using speech classi cation to match the EEG data. Further, two di erent calibrationmethods are tested: additional mini-training sets for each person, and a group of normalized training sets, andthe results are presented.