Music is vital for entertainment, emotion regulation, and stress relief. With digital platforms like Spotify, classifying large music datasets has become essential. This introduces a machine-learning framework to detect four emotions Happy, Sad, Calm, and Energetic in Hindi songs. Music has the unique ability to evoke and convey a wide range of human emotions, making it a powerful medium for both artistic expression and practical applications. A curated Hindi music dataset was segmented into 20-second WAV clips (44.1 kHz), preprocessed with high-pass filtering and volume normalization. Acoustic features extracted included: (1) 13-dimensional MFCCs, (2) 12-dimensional chroma vectors, (3) Zero-Crossing Rate, and (4) Spectral Rolloff. Data was split into training (70%), validation (15%), and testing (15%) sets using stratified sampling. Three classifiers were applied: Decision Tree (max depth 10), Random Forest (100 trees, depth 12), and XGBoost (200 estimators, learning rate 0.1, depth 6). XGBoost performed best with 86.4% accuracy, while Random Forest and Decision Tree achieved 83.6% and 74.2%, respectively. "Sad" and "Calm" were the most confused classes (~8%). Results show ensemble models effectively classify emotions in regional music and support applications like mood-based playlists and smart music systems.
Shelke et al. (Fri,) studied this question.