A Comparative Study on MFCC and Fundamental Frequency Based Speech Emotion Classification

Abstract

Speech emotion recognition and classification is one of the most important and emerging fields in artificial intelligence. It has various uses in different applications starting from medical science to smart home devices. Input feature selection is a very important part of speech processing. Mel Frequency Cepstral Coefficients is the most widely used features in the processing of audio data. In case of processing of emotion related data, the fundamental frequency also plays an important role. In this study a comparative analysis has been conducted to determine the better feature in the field of emotion classification. Emo-Db database was used for the study. For classification task the Support Vector Machine classifier with the radial basis and sigmoid function kernel has been used. The model was trained with both the audio features and the performances were compared. Better performance was observed with Mel Frequency Cepstral Coefficients which ensures the better performing speech features in emotion classification task.