A Visual Interface for Emotion based Music Navigation using Subjective and Objective Measures of Emotion Perception

Abstract

Beyond the conventional playlists organized based on genre, artist, or album, the study attempts to visualize a music library on a 2D plane enabling the users for emotion based navigation. A dataset constructed for this research purpose, comprising of music stimuli which associate Sri Lankan folk melodies has been utilized in the study. Using the method miremotion defined in MATLAB MIRToolbox, the objective predictions for the discrete emotions/emotion dimensions have been obtained. Subjective emotion ratings were collected through a listening experiment. By running a Principal Component Analysis on the data matrix combining subjective and objective measures, music library is mapped on to a 2D plane while identifying the correlations among each measures. Several directions have been identified for extending the research whereas the cultural specificity in emotional expression in music and the subjectivity in emotion perception have made the emotion based music recommendation a challenging task.