Emotional speech recognition from an aud

The main aim of this study is to recognise emotions from auditory stream. This study also aims to compare different feature selection techniques and compares results of different intelligent techniques when applied to classify different number of features. To achieve the goals of the study, existing literature available in the field of emotion recognition, signal processing, data mining and intelligent techniques was reviewed. Based on the knowledge gained by reviewing the literature, a viable methodology was designed to implement the experiments required to achieve the goals of this project. The methodology was designed on the basis of common approach followed by many studies, which included selection of an emotional database, extraction of features from the emotional database, reducing the feature set by using a ranking algorithm and finally processing the features through intelligent techniques to classify the features into different emotions. The methodology was implemented using OpenSMILE and Weka for feature generation and feature reduction respectively, from the Berlin Emotional Database. The obtained 988 features were reduced using SVMAttributeEval algorithm in Weka, then processed and classified using two intelligent techniques SVM and MLP as required by the study. Given the current popularity of speech recognition techniques in day-to-day use, the use of emotional recognition from speech will play a vital role in further development of this technology and drastically improve the quality of human-computer interaction through speech.