Automatic Genre Classification for Sri Lankan Music with Machine Learning Approaches

Abstract

The music of Sri Lanka has originated from different influences such as; ancient folk rituals, Buddhist religious traditions, the legacy of European colonization, and the commercial and historical influence of nearby Indian culture. But there was a need of creating a Sri Lankan musical culture based on our own Sri Lankan Tradition. In 1940s, the musician Sunil Santha created a landmark in the musical arena of Sri Lanka. Following his path, many Sri Lankan musicians attempted to create music which has an unique Sri Lankan Identity. As a result of that currently we can find set of music compositions which contains the unique features of the music of Sri Lanka. The main focus of this research is to find out set of features which describes the unique identity of Sri Lankan music & to build a classifier using MFCC features extracted from predominant melodic pitch contours, estimated from polyphonic music audio signals. The classifier focuses on differentiating between distinct singing styles of Sri Lankan folk & other styles such as Indian Classical & Western Classical. It uses existing technology for feature extraction, feature selection & classification. Different Standard Classification algorithms will be compared to automatically distinguish difference between Sri Lankan music from other music styles, using the extracted features. Finally, the model is evaluated and refined, and a working prototype is implemented.