Classification of short-duration sounds for environmental monitoring

In this study, we are interested in the classification of short-duration sounds related to surveillance context. We carefully select a set of features allowing a better discrimination of the signals. Considering each pattern vector, we introduce the mean and standard deviation of every feature components. We also explore the way the signal is more appropriately analyzed by considering possible partitioning into three segments of the signal. The classification is performed by an SVM classifier implemented using the SMO algorithm. We note that adding the standard deviation improve the classification performance rate for this type of sounds. Experiments present various results concerning the signal partitioning. They show that partitioning does not enhance the classifier performance.