Hierarchical Support Vector Machines for Audio Classification

Abstract

Audio \u00a0data \u00a0is \u00a0one \u00a0of \u00a0typical \u00a0multimedia \u00a0data \u00a0and \u00a0it \u00a0contains \u00a0plenty \u00a0of \u00a0information. \u00a0Audio retrieval \u00a0is \u00a0becoming \u00a0important \u00a0content \u00a0in \u00a0multimedia \u00a0information \u00a0retrieval. \u00a0In \u00a0multimedia \u00a0retrieval researches, it \u00a0becomes \u00a0more \u00a0and \u00a0more important research part how \u00a0to \u00a0construct \u00a0better \u00a0classifiers \u00a0for \u00a0audio classification and retrieval. \u00a0Support \u00a0Vector \u00a0Machines, \u00a0a novel \u00a0method \u00a0of \u00a0the \u00a0Pattern \u00a0Recognition, \u00a0presents excellent \u00a0performance \u00a0in \u00a0solving \u00a0the \u00a0problems \u00a0with \u00a0small \u00a0sample, \u00a0nonlinear \u00a0and \u00a0local \u00a0minima. \u00a0But \u00a0audio classification \u00a0is \u00a0a \u00a0multi-class \u00a0classification \u00a0problem \u00a0and \u00a0it(cid:146)s \u00a0just \u00a0one \u00a0of \u00a0problems \u00a0to \u00a0be \u00a0solved \u00a0in \u00a0SVM researches. In this paper, it compares several common Support Vector Machines and proposes a hierarchical Support Vector Machines based on audio features cluster method, combining audio features and hierarchical SVMS. It uses hierarchical classification method to classify audio data and it(cid:146)s proved better performance by experiments.