Significance of prosodic features for automatic emotion recognition

This paper aims to recognize emotions from speech on a more realistic database using various classifiers. For this purpose, experiments are conducted using the standard 6373 dimensional Computational Paralinguistic Challenge (ComParE) feature set. The features extracted are modeled using Support Vector Machine (SVM) and Deep Neural Network (DNN) classifiers. The effectiveness of the proposed system has been validated on the Emotional Sensitivity Assistance System for People with Disabilities (EmotAsS) database, provided as part of the INTERSPEECH 2018 Computational Paralinguistics Challenge. Besides, experiments have also been performed on a reduced subset of the standard ComParE acoustic feature set consisting of 873 prosodic features. Experimental results suggest that the reduced prosodic feature set provides comparable performance with the original feature set. It is also observed that DNN classifier provides better performance than SVM.