Study of Substitute Vocalization System based on the Surface Electromyogram

本研究的目的是在不能进行有声语音交流的情况下，利用面部肌肉的表面肌电信号（SEMG）来实现无声语音识别。然而，对于一些初学清音语音的人来说，识别准确率较低。为了解决这个问题，我们研究了使用自组织映射（SOM）算法从多个扬声器测量的活动值的分类，并确定了每个元音的活动值的标准模式。此外，作者还提出了一个清音训练系统。

The purpose of this work is to achieve unvoiced speech recognition using Surface Electromyogram (SEMG) of the facial muscles when voiced communication is impossible. However, the recognition accuracy was low for some speakers who were beginner of unvoiced speech. To address this issue, we have examined the classification of the activity values measured from multiple speakers using Self-Organizing Map (SOM) algorithm and determined the standard pattern of the activity values for each vowel. In addition, the author has proposed a training system for unvoiced speech.