Latent Dirichlet Allocation Model for Recognizing Emotion from Music

Abstract

The recognition of emotion has become a multi-disciplinary research area that has received great interest. Recognizing emotion of audio data will be useful for content-based searching, mood detection etc. The goal of this paper is to elaborate a system that automatically recognizes the emotion of the music. We present a technique used for document classification, Latent Dirichlet allocation (LDA) for the purpose of identifying emotion from music. The recognition process consists of three steps. In the first step, extractions of ten distinct features from music are performed followed by Clustering of values of these features, and finally in the third step an LDA model for each of the emotions is constructed. After constructing the LDA the emotion of the given music is identified. This model was tested on South Indian film music to recognize 6 emotions happy, sad, angry, love, disgust, fear and achieved an average accuracy of 80%.