Using One-Class SVMs and MP for Audio Recognition of Action Scenes

The paper presents a system for sounds classification of action scenes. We propose to use the matching pursuit (MP) algorithm to obtain effective time-frequency features. The MP-based method utilizes a dictionary of atoms for feature selection, resulting in a flexible, intuitive and physically interpretable set of features. Then apply one-class support vector machines (1-SVMs) together with MP features and MFCC to classify the sounds. The paper considers six type sounds: the sword sound, the club sound, the unarmed sound, the broken sound, the metal-falling sound and the shout sound. Experimental results prove that the method is effective.