Pattern Recognition of Facial Electromyography (FEMG) Signal for Aceh Language Speech using Naïve Bayes and Learning Vector Quantization (LVQ)

Abstract

Facial electromyography is myoelectric signals that formed by human facial muscles. This signal can be acquired by attaching an electrode to the facial muscle that has been connected with an electromyography sensor. When human say certain words, the articulation muscles contract and facial electromyography signals appear in the muscles. This study aims to recognize patterns in facial electromyography signals by classifying signals using naïve bayes and learning vector quantization classifier. Feature extraction used one-dimensional discrete wavelet transforms. Wavelet transform type used wavelet daubechies2 level 5. The transformation produces a level 5 approximation coefficient called a5 and five detail coefficients called d1, d2, d3, d4, and d5. The result of this study show that the average classification accuracy for ho neuk jak sentence using naïve bayes and LVQ classifier was 62.5% and 92.5% respectively. The average classification accuracy for ja’ word using naïve bayes and LVQ classifier was 70% and 92.5% respectively. The average classification accuracy for ja’ wo sentence using naïve bayes and LVQ classifier was 52.5% and 90% respectively. The average classification accuracy for pane word using naïve bayes and LVQ was 70% and 90% respectively. The average classification accuracy for soe word using naïve bayes and LVQ classifier is 85% and 95% respectively. Thus, this study shows that when humans say the words, facial electromyography signals that appear on facial muscles difference for each subject.