Sub-vocal phoneme-based EMG pattern recognition and its application in diagnosis

Sub-vocal speech recognition provides Silent speech communication that does not disturb or interfere with the surrounding, thus confidential information can be submitted securely. Sub-vocal speech depends on the brain´s normal output of peripheral nerves and muscles. In this paper, we introduce a new method for statistical features extraction to detect and classify the for sub-vocal EMG pattern. This method is based on exploring the both time and frequency domain information to explore the features. Discrete Wavelet Transform (DWT) is used for signal processing and sub-band feature extraction. Two features Mean absolute deviation and standard deviation have been extracted from different sub-band. Then, these features were fed into a linear classifier to classify the pattern of sub-vocal phonemes. Results showed that accuracy, sensitivity and specificity rates ranges from 70% to 80% is achieved, hence can be used to design expert system for the application in diagnosis purposes.