文献类型：期刊文献

中文题名：基于EMG信号的无声语音识别应用及实现

英文题名：Application and Realization of EMG in Unvoiced Speech Recognition

作者：许佳佳[1];姚晓东[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2006

卷号：34

期号：5

起止页码：133

中文期刊名：计算机与数字工程

外文期刊名：Computer & Digital Engineering

收录：CSTPCD

语种：中文

中文关键词：EMG信号;语音识别;小波变换;神经网络

外文关键词：electromyography signal, speech recognition, wavelet transform, neural network

摘要：提出了基于肌电信号(EMG)的无声语音识别系统。由于该系统是通过EMG信号而非声音信号进行识别,因此可应用于高噪声环境和帮助失去发音能力的人实现无声交流,有着良好的应用前景。关于该系统的实现,提出了以下方法:实验时使用0-9十个中文数字,由受试者不发声地重复说出,从三块面部肌肉采集EMG信号;对EMG信号进行小波变换,获取变换系数矩阵后提取其能量值,构造特征矢量送入BP神经网络分类器分类。实验表明,基于小波变换的特征提取方法是一种有效的方法,适用于类似EMG信号的非平稳生理信号。
It is proposed in this paper that electromyography（EMG） signal can be used in unvoiced speech recognition systems. It recognizes speech by observing the facial muscles associated with speech. Because voice signals are not used, it can be used in noisy environments and also supports people with speaking diffculties. Experiment is performed using a tenword vocabulary consisting of the numbers ＂zero＂ to ＂nine＂, subjects repeated each word without any voice. EMG signals from three muscles are processed using Wavelet Tramform; the wavelet coefficients are inputed to a neural network. The recognition accuracy is over 95%. The results demonstrate that there is excellent potential for using EMG to enhance the performance of a conventional speech recognition system and the realization method proposed in this paper is efficient for processing EMG.