MF143 表情筋および頸部の筋電位信号を利用した日本語母音の識別(MF14 情報・コミュニケーション支援,あたり前のことを知る)

Abstract

The purpose of this paper is to achieve Japanese vowels classification without using speech signals. For this purpose, we propose a method for classifying Japanese vowels using EMG signals with Support Vector Machine (SVM). Especially, we focused on vowel classification using EMG signals measured from neck only. Classification experiments for 5 Japanese vowels were conducted with four subjects by using our proposed method. As the experimental results, three subjects showed the classification accuracy of more than 80% using EMG signals measured from facial muscles and neck. When using EMG signals measured from neck only, the classification accuracy was decreased about 17-30% compared to from facial and neck. However, two subjects showed the classification accuracy more than 60%.