Hierarchical Multi-class Audio Classification using Support Vector Machine

Abstract

Automatic audio classification is an important part in audio indexing, content-based audio retrieval and online audio distribution. The efficiency of an audio classification or categorization depends on the ability to capture proper audio features and to accurately classify each feature set corresponding to its own class. In this paper a novel hierarchical audio classification algorithm using Support Vector Machine(SVM) is proposed to classify the  input digital audio stream into five audio classes namely male speech,female speech, music, silence and environmental sound. An audio database “Swarsanchay” consists of recordings of 1000 audio clips of all the above mentioned audio classes with different noisy conditions was created for experimentation. Out of which five hundred clips were used for training and remaining five hundred audio clips were used for testing. The proposed multi class audio classifier correctly identified 475 audio clips giving an overall classification accuracy of 95%. During our investigation of feature extraction we identified best ranking features pertinent to class having better discriminatory power amongst the classes. The best classification accuracy of 95% is obtained when best ranking feature set is used for hierarchical multiclass classification.