A Hybrid Computational Framework Using NLP and ML for Emotion Analysis in Sinhala Songs

This study Rasathura explores the emotional dimensions of Sri Lankan music, specifically Sinhala songs, by developing an emotional classification model using both audio features and lyrical analysis. The research addresses the limitations in emotional classification using advanced Natural Language Processing (NLP) techniques for Sinhala music. For audio analysis, JAudio was used to extract musical data, and machine learning (ML) techniques, including Random Forest, Sequential Minimal Optimization (SMO), Naive Bayes, Decision Tree, and Logistic Regression, were evaluated for classification. Relief-based and correlation-based attribute selection were utilized to enhance classification accuracy. The SMO algorithm achieved an accuracy of 75.92% using JAudio features, which increased to 79.01% with Relief-based attribute selection. The BERT model (built on XLM-RoBERTa) was used for lyrical analysis, achieving a maximum accuracy of 61.22%. TFIDF, Word2Vec with n-grams, and BERT embeddings were evaluated for feature extraction in lyrical data. The findings could benefit the future music industry by creating personalized music recommendation systems, enhancing music therapy applications, and contributing to preserving cultural heritage through digital music archiving.