EMOTION RECOGNITION OF MUSIC BY USING CORRELATION BASED ANNOTATION PROCESS

Dr. Deepti Chaudhary

Abstract

The boost up in the use of internet and vibrant lifestyle provoked the use of automated music players. The emotion identification is the foremost task for automated music players. It provides the solutions to the problems that come into account while exploring and downloading the songs of particular category from largely available online data. The emotion perception varies widely from person to person. The emotion recognition system basically consists of five steps that include dataset collection, preprocessing, annotation, feature extraction and classification. In the proposed work the music annotation step is explored by using correlation based approach (CBA) and is compared with human defined annotation (HDA). CBA is algorithm based and HDA is subjective. The database considered for the research consists of Hindi songs and GTZAN standard dataset for research. In the proposed approach correlation values of the feature vectors of songs belonging to different categories are used to train the support vector machine (SVM) instead of subjective classification carried out by group of people as in HDA. Based on the analysis conducted the CBA outperforms HDA in terms of evaluation parameters considered i.e. accuracy, recall and precision. The accuracy achieved by using CBA for Hindi songs is 90.2% and for GTZAN dataset is 86.3%.