Gender specific emotion recognition through speech signals

This paper proposes an emotion recognition system which allows recognizing a person´s emotional state from speech signal. The aim of proposed solution is to improve the interaction among humans and computers. The emotion recognition system must be capable of recognizing at least six basic emotions (happiness, anger, surprise, disgust, fear, sadness) and the neutral circumstances. The proposed system has two subsystems Gender Recognition (GR) and Emotion Recognition (ER) and also distinguishes a single emotion versus all the others. An appropriate emotion recognition method is applied after extracting features like pitch, energy and MFCC having emotional information. The performance in terms of accuracy is shown in results. The highlight of result is that a prior knowledge about the gender of speaker increases the performance of proposed system. Proposed approach has been implemented by using Naive Bayes method. This is a simple and efficient classification approach. It has easy learning on large speech databases and its accuracy as compared to other approaches is reasonably good. In future this system can be implemented over mobile devices such as smart phones.