Acoustic Events Classification Using Support Vector Machines (SVMs)

Abstract

In this paper, an approach is built to automatically detect acoustic events that are produced in a meeting or lecture room environment. Six audio classes are to be classified through this approach. The classes considered are music, speech, clapping, door slam, cough, and laughter. Several events samples are collected from the Internet. Support Vector Machines (SVMs) perform training and testing the events classification on perceptual and MFCC features set. A hierarchical clustering scheme is used therefore the required number of binary SVM classifier is also reduced. The system is tested on different data sets and its effectiveness is determined with classification accuracy on audio event frames.