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

Abstract

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.

Automated Transcription for Raga Recognition and Classification in Indian Classical Music Using Machine Learning

“…But for amplitude-based features limited the performance of the model and obtained accuracy ranging from 49 to 63% and Standard deviation of 0.037 The developed model results from [9] were obtained based on characteristics of the dataset-required size, class balance, quality of the annotations required improvement for achieving good performance in terms of accuracy. The results obtained for the model in terms of accuracy was of 75% and Fmeasure of 73 % In [11] developed model obtained results based on the pattern recognition among the obtained pitches an accuracy 94% and Error value of 6. In [12] developed a model that utilized various classifiers for raga recognition distinguished lowered more information and resulted in over-fitting issues.…”

“…However, the characteristics of the dataset-required size, class balance, quality of the annotations were still more needed to improve for achieving good performance in terms of accuracy. Kaur and Kumar [11] developed Mean Centered Clustering (MCC) that was utilized for tune grouping for automated raga recognition. The created MCC procedure boosted the distance among the tones and decreased the spread of information in singular groups.…”