Recognition of Imagined Speech using Electroencephalogram Signals

Mind reading, or imagined speech recognition, has many real-world applications including silent communication between soldiers in a combat zone, lie-detectors, and even communication with stroke victims who are fully coherent but unable to produce verbal speech. However, while previous research has shown that using electroencephalography (EEG) to read imagined speech is certainly possible, it is currently impractical because each EEG device can be calibrated for only one person. No calibration test has been developed to allow the same EEG device to accurately read the imagined speech of different people. A calibration test is necessary because different people have different neurons firing even when thinking of the same letter, word, syllable, etc. For example, the neurons that fire when person A thinks of the letter 'a' is different than those fired when person B thinks of the same letter. In this paper, and as a first step we demonstrate the possibility of reading a brain using noninvasive (EEG) technology. Further an algorithm is developed to calibrate the same device for different subjects. This is accomplished first by gathering raw EEG data, performing pre-processing on that data, performing feature extraction, and finally using speech classification to match the EEG data. Further, two different calibration methods are tested: additional mini-training sets for each person, and a group of normalized training sets, and the results are presented.