基于内容音频检索中boosting算法.pdfVIP

BOOSTING FOR CONTENT-BASED AUDIO CLASSIFICATION AND RETRIEVAL: AN EVALUATION Guodong Guo, Hong-Jiang Zhang, and Stan Z. Li Microsoft Research China 3F, Beijing Sigma Center, No. 49, Zhichun Road, Haidian District Beijing 100080, P.R.China ABSTRACT probability of misclassifying yet-to-be-seen patterns for a fixed but unknown probability distribution of the data. This In this paper, we evaluate a recently proposed algorithm in is in contrast to traditional pattern recognition techniques machine learning called AdaBoost for content-based audio of minimizing the empirical risk, that is, optimizing the per- classification and retrieval. AdaBoost is a kind of large mar- formance on the training data. This minimum structural risk gin classifiers and is efficient for on-line learning. Our focus principle is equivalent to minimizing an upper bound on the is to evaluate its classification and retrieval accuracy as com- generalization error. pared with other methods. The Muscle Fish audio database Boosting [7] also tries to maximize the margin between of 409 sounds is used for the evaluation with perceptual and positive and negative examples, in which ensemble of clas- cepstral features. sifiers are trained sequentially. In