Machine Learning for Emotion Classification of Sri Lankan Folk Music
Joseph Charles, Sugeeswari Lekamge · 2019
Music conveys and evokes powerful emotions owing to various musical characteristics such as rhythm, melody, and orchestration. This amazing ability has motivated the researchers worldwide to discover relationships between music and emotion. As a result, various data mining tasks have been carried out where state-of-the-art machine learning techniques are utilized in music emotion classification. However, the literature reveals that these studies frequently utilize western or western classical music whereas no considerable effort is reported in computational modeling of cultural-specific music e.g., Sri Lankan folk melodies, despite being an abundant source of emotion expression. Further, the applicability of existing classifiers trained using different ground-truth data, in other cultural-specific content is found to be problematic. Therefore, we applied machine learning techniques for emotion classification of a dataset which associates Sri Lankan folk melodies. Using MATLAB MIRToolbox, acoustic features pertaining to dynamics, rhythm, timbre, pitch, and tonality were extracted. Performance of five standard classification algorithms (Support Vector Machine, Naive Bayes, Decision Tree, Random Forest, and k-Nearest Neighbor) were tested on the dataset comprising of 206 music stimuli (30s; 44100Hz; stereo; 32bit;. wav) representing happy, sad, and fear as the predominant emotions. Among the classifiers, k-NN yielded the highest accuracy value of 78.44%. Moreover, highly influential features were identified based on the correlations between the acoustic features and the class labels which resulted in an enhanced accuracy of 87.95% with k-NN. The findings of the study mark a promising initial step, introducing machine learning for emotion analysis of Sri Lankan folk melodies.