Audio Classification Framework based on DSVM

Abstract

The explanation of how a decision made is important for accepting the machine learning technology, especially for applications such as multimedia. SVMs have shown strong generalization ability in a number of application areas, including audio classification. However, the poor comprehensibility hinders the success of the SVM for audio class prediction. On the other hand, a decision tree has good comprehensibility. This approach combines the SVM with decision tree. We use the SVM as a decision of binary tree to select strong instances to generate rules; it is known as decision SVM (DSVM). We considered eight audio classes: silence, non-silence, speech with music, speech with noise, music with environment sound, instrumental music, environmental sound and noise. This classification and analysis is intended to analyze the structure of the sports video.