This paper presents a method of identifying human emotions through spoken conversations. The proposed work involves the use of English language real-time database that we have created and various techniques of feature extraction, such as pitch frequency, formant frequency, energy and MFCC. In this speech based technique, a lot of data can be found, then size of feature is deduced using Principle Component Analysis (PCA) technique and the performance is evaluated on the basis of emotion classification accuracy. We evaluated the results using accuracy, sensitivity and specificity parameters for SVM based classifier results. The accuracy up to 60% is achieved with PCA and 53% is achieved without PCA for detection of emotion.

real-time database, feature extraction, principal component analysis (PCA), support vector machine (SVM) classifier.