Identification of Most Contributing Features for Audio Classification

Audio classification is very essential for faster retrieval of audio files. Extracting best set of features and deciding best analysis method is very important for getting best results of audio classification. In this paper, we have used distinct feature selection methods to identify the most relevant and non-redundant feature set for audio classification into four classes: pure speech, pure music, silence and noise. With these set of features and Support Vector Machine (SVM) as a classifier we have got the precision of 99.8% and recall of 99.9%, which is more promising than the previous approaches.