Human speech sentiments recognition: A data mining approach for categorization of speech

This paper proposes gender dependent emotion recognition system. We have collected a huge database that consists of emotional speech generated by males and females under different age groups. Based upon the speech produced by different people, each entity speech is then classified into different categories such as happy, sad, anger, bored and neutral. Various features extraction techniques is applied to extract the features from speech. The classification techniques such as support vector machine, linear discriminant analysis classifier, K-nearest neighbors and naive Bayes classifier are then applied on the extracted features to classify the emotions. A gender classification system is adopted based on Gaussian Mixture Model for classifying the gender. Results show that the accuracy of emotion classification system is above 99 % if gender is known.