Sign language-Thai alphabet conversion based on Electromyogram (EMG)

Abstract

Communication and sign-language learning of the people with hearing disabilities in Thailand has been problematic due to limited number of sign-language experts. To facilitate the sign-language learning and communication between the hearing disability and ordinary people, the sign language-to-alphabet spelling conversion was developed based on electromyography (EMG) signal recorded from the forearm muscles. The EMG signal of 10 different Thai sign-language gestures were recorded with the electrode arrangement similar to the Myo device from Thalmic Labs and analyzed. To extract the distinct features of the EMG signals, moving variance and mean absolute value (MAV) were chosen. The extracted output data was processed with the classification algorithm via non-linear model (artificial neural networks (ANN)) to confirm that the EMG signal for each alphabet gesture is accurately matched with the actual spelling alphabet. The system is able to measure the match of the output with total accuracy of more than 95%.