基于小波特征和多类支持向量机的病态语音识别方法

Application of modified wavelet features and multi-class SVM to pathological vocal detection

研究一种应用小波特征向量和多类支持向量机进行病态语音识别的方法,该方法基于连续小波变换提取语音特征向量,利用多类支持向量机进行病态语音分类。为了简化二分类支持向量机进行多类分类时所带来的计算复杂性,根据一类支持向量机分类思想提出一种多类分类算法。该算法能够使每一类样本都独立地获得一个决策函数,通过决策函数的最大值来判断样本所属的类。实验表明,在病态语音识别系统中,多类支持向量机与小波特征向量相结合具有良好的识别效果和应用价值。

This paper researched the method of wavelet feature-vectors and multi-class Support Vector Machines （SVM） applied to pathological vocal detection, which extracted features of the pathological vocal based on continuous wavelet transformation and then classifies pathological vocal by multi-class support vector machine. In order to reduce computation complexity caused by using the standard SVM for multi-class classification, a new multi-class classification algorithm based on one-class classification was proposed. It can form a decision function for every single class sample and accordingly obtain the aim of classification based on maximum of decision function. Experimental results have shown that the pathological vocal detection system is feasible and applicable by the combination of multi-class SVM and wavelet feature-vectors.