Music Emotion Recognition in Assamese Songs using MFCC Features and MLP Classifier

Abstract

Music Emotion Recognition (MER) is one of the fastest growing research topics and important subfield of Music Information Retrieval (MIR) system that has grown in recent years to improve Human Machine Interaction (HMI). A tremendous research is being done on high-resourced languages like English, whereas very less work has been performed in music emotion recognition on Assamese (a regional language from North-Eastern India) songs. This paper attempts to perform a novel and simple solution to the problem of emotion recognition in Assamese songs. We used a newly created Database of Assamese songs ASDB consisting of 80 samples using eight well known Assamese singers. The performance of MER with MFCC and chroma features has been analyzed in this work. Multi-Layer Perceptron (MLP) Classifier is used for emotion recognition. By analyzing the results, it is found that MFCC can give better MER accuracy of 93.75% than the chroma features with the input audio sample of length 4 seconds.