kNN Ensemble for Karnatik Raga Classification
Author(s):
Vinuraj Devaraj , Harrisburg University of Science & Technology; Anjana Suresh, Mphasis Ltd.Keywords:
Voting Classifier, kNNAbstract:
In this paper we propose a soft Voting Classifier ensemble of k-Nearest Neighbor (kNN) models to classify Karnatik music sound signals into one of six Karnatik ragas viz. Bhairavi, Kalyani, Khamboji, Mohanam, Shankarabharanam and Todi. The proposed architecture achieves a test accuracy of 91% on sound data containing noise with maximum class validation accuracy of 93.2% (for raga Khamboji) and minimum class validation accuracy of 90.4% (for raga Todi). We also observe that, sampling sounds, especially musical sounds, at intervals of 10 seconds or 20 seconds might yield better model accuracies as compared to sampling at level higher or lower than the specified duration, especially when the sound signals contain noise.
Other Details:
| Manuscript Id | : | J4RV6I11001 |
| Published in | : | Volume : 6, Issue : 11 |
| Publication Date | : | 01/02/2021 |
| Page(s) | : | 1-6 |
