Comparison of Artificial Neural Network and Gaussian Mixture Model Based Machine Learning Techniques Using DDMFCC Vectors for Emotion Recognition in Kannada

Abstract

We build an emotion recognition system based on Artificial Neural Network (ANN) and compare the same with the one based upon the Gaussian Mixture Modeling (GMM) scheme. Both the systems were built upon probabilistic pattern recognition and acoustic phonetic modelling approaches. Since our native language is Kannada, one of the very rich Indian language, we have used words uttered in Kannada to train and test the schemes. Since Mel Frequency Cepstral Coefficients (MFCC) are well known acoustic features of speech [1] [2] [4], we have used the Delta MFCC and the Double Delta MFCC vectors in speech feature extraction. Finally, performance analysis of these models in terms of Emotion Error Rate (EER) justifies the fact that modeling using the ANN yields better results over other modeling schemes and can be used in developing Automatic Emotion Recognition systems.