一种基于ELM与LDA的音频信号分类算法

摘要

针对多媒体信息中的音频信号,提出一种基于线性判别分析(LDA)与极限学习机(ELM)的分类方法.首先,使用傅里叶变换等方法从每一段音频中提取特征,并将它们按比例组成一个高维向量；其次,应用LDA对高维向量进行降维,使其成为用于分类的最优特征,作为ELM的训练和测试样本；最后,分别采用ELM,SVM,BP分类器对4种音频信号进行分类,并进行性能对比与分析.实验表明:提出的算法对于较难分的类也具有较好的分类效果,平均正确率为90％,同时运算速度比SVM快一千多倍.%In this paper,a new classification method is designed based on linear discriminant analysis (LDA) and extreme learning machine (ELM) for audio signals of Multimedia information.We first apply methods such as Fourier transform to extract features from each section of audio signal,which would be proportionally organized to form a high-dimensional vector.And then LDA method is applied to reduce the dimensionality of the feature vector for making it the best feature vector for classification,which can be the training and testing sample of ELM.Finally,use ELM,SVM,and BP classifier are used to do experiments on four kinds of audio signal respectively,and their performances are contrasted.The result shows that,the method promoted still exhibits better classification performance for the classes which are hard to classified,with average accuracy rate of 91％,and its computing speed is more than one thousand times faster than SVM.