Emotion-Based Hindi Music Classification

Abstract

Music emotion detection is becoming a vast and challenging field of research with the increase in digital music clips available online. Emotion can be considered as energy that brings the person in positive or negative motion. Music emotion recognition (MER) is an emerging field of research in various areas such as engineering, medical, psychology, musicology. In this work, the music signals are divided into four categories—excited, calm, sad and frustrated. Feature extraction is carried out by using MIR toolbox. The classification is done by using K-nearest neighbor (K-NN) and support vector machine (SVM). The feature-wise accuracy of both the approaches is compared by considering mean and standard deviation of feature vectors. Results reveal that spectral features provide the maximum accuracy among all the features and SVM outperforms K-NN. Maximum accuracy achieved by SVM is 72.5% and K-NN is 71.8% for spectral features. If all the features are considered, then the accuracy achieved by SVM is 75% and by K-NN is 73.8%.