An Incremental Support Vector Machine based Speech Activity Detection Algorithm

Traditional voice activity detection algorithms are mostly threshold-based or statistical model-based. All those methods are absent of the ability to react quickly to variations of environments. This paper describes an incremental SVM (support vector machine) method for speech activity detection. The proposed incremental procedure makes it adaptive to variation of environments and the special construction of incremental training data set decreases computing consumption effectively. Experiments results demonstrated its higher end point detection accuracy. Further work will be focused on decreasing computing consumption and importing multi-class SVM classifiers