Mean centred clustering: improving melody classification using time- and frequency-domain supervised clustering

This paper reports a new approach for clustering melodies in audio music collections of both western as well as Indian background and its application to genre classification. A simple yet effective new classification technique called mean centred clustering (MCC) is discussed. The proposed technique maximizes the distance between different clusters and reduces the spread of data in individual clusters. The use of MCC as a preprocessing technique for conventional classifiers like artificial neural network (ANN) and support vector machine (SVM) is also demonstrated. It is observed that the MCC-based classifier outperforms the classifiers based on conventional techniques such as Principal Component Analysis (PCA) and discrete cosine transform (DCT). Extensive simulation results obtained on different data sets of western genre (ISMIR) and classical Indian ragas are used to validate the efficiency of proposed MCC-based clustering algorithm and ANN/SVM classifiers based on MCC. As an additional endeavour, the performance of MCC on preprocessed data from PCA and DCT is studied. Based on simulation results, it is concluded that the application of MCC on DCT coefficients resulted in the highest overall classification success rate over different architectures of the classifiers.