Unvoiced Chinese Digital Recognition Based On Facial Myoelectric Signal

Different from multi-channel MES, eleven Chinese digitals zero to ten (/i/, /er/, /san/, /si/, /wu/, /liu/, /qi/, /ba/, /jiu/, and /shi/) are studied based on the one-channel detected myoelectric signal (MES). Zygomaticus major and anterior belly of the digastric are carefully selected as the electrodes site of MES detected. According to MES characteristic, wavelet transform coefficients, AR model coefficients and real cepstral coefficients are calculated as the features of MES. Using GA (genetic arithmetic) sixteen features are selected from the original features as the inputs of SVM (support vector machine) classifier. The result shows that using the MES to recognize unvoiced speech is a promising way