EMD近似熵与支持向量机在音频分类处理中的应用

Application of EMD Approximate Entropy and Support Vector Machine Method in Processing Audio Classification

将动物声音作为音频源,针对家养动物声音的非线性、非平稳特征和在现实条件下难以获取大量动物声音样本的实际情况,提出一种经验模态分解(EMD)近似熵(ApEn)结合支持向量机(SVM)的家养动物声音分类识别方法.通过EMD方法将非平稳的动物声音信号分解成若干个平稳的固有模态函数(IMF);对IMF进行筛选,计算所筛选IMF的近似熵构成特征向量;将特征向量输入SVM分类器进行分类识别.对家养动物声音样本按该方法进行测试,结果表明,该方法能有效提取声音特征,在小样本情况下也具有较高的精度和较强的泛化能力,该方法能有效地应用于动物声音的识别分类.

In this paper, animal sounds are used as the audio source. According to the non-stationary and non-linear characteristics of domesticated animal voice and the situation in which it＇s hard to obtain enough sound samples, a domesticated animal voice diagnosis method based on Empirical Mode Decomposition （EMD）, Approximate Entropy（ApEn）,and Support Vector Machine（SVM） is proposed. Firstly, the domesticated animal signals are decomposed into a finite number of intrinsic mode function（IMF）. Then, the ApEns of five IMFs filtered are used to form eigenvectors. Finally, the eigenvectors are put into a support vector machine categorizer. The results of animal data experimental recognition show that this method has high accuracy and good generalization abilities even in the case of small number of samples. The approach proposed can identify the domesticated animal voice effectively.