Raga identification using clustering algorithm

The main objective of this paper is to evaluate the performance of raga identification using clustering algorithm and to compare the accuracy of the raga identification system which uses Mel Frequency Cepstral Coefficients (MFCC) as feature and the system which uses pitch information along with MFCC features. The goal of Raga identification is to identify the raga independent of training data. The training and testing phase are done for direct film song (vocal with background music) for 10 ragas. In training phase 10 film songs for each raga based on singers is taken as input. The input songs are made to undergo frame blocking. The Mel Frequency Cepstral Coefficients (MFCC) are extracted for each frames of the input signal. Similarly training is also done with MFCC and Pitch features (by Cepstrum method). The raga model is developed by K-means clustering algorithm for each raga. In clustering method, the cluster centroids are obtained for cluster size of 256 and stored. One model is created for each raga. In the testing phase Minimum mean of distances is computed for each model. Raga is classified based on selection of the model which produces minimum of average. The main motive behind Raga identification is that it can be used as a good basis for music information retrieval of any Carnatic music songs or Film songs.