According to the nonlinear characteristics of human vocal system and the suppress chaos characteristics of speech signal, a method of speech signal feature extraction which based on the nonlinear resonances Duffing model is proposed. The results show that the feature presented has higher recognition rate compared to the generalized dimensions feature based on nonlinear dynamics, meanwhile, the same speech signal has a higher recognition rate than the classical algorithm of MFCC feature.%针对人发声系统的非线性特性和语音信号的类混沌特性,提出了一种基于非线性共振Duffing模型的说话人语音信号的特征提取方法.实验结果表明:采用非线性共振Duffing模型的特征提取方法,较基于非线性动力学提取广义维数特征具有较高的识别率.同时,同一语音信号在相同的识别系统中,与经典的MFCC特征相比,也具有较高的识别率.
展开▼