Classification of Emotional Speech of Children Using Probabilistic Neural Network

Abstrak

Child emotions are highly flexible and overlapping. The recognition is a difficult task when single emotion conveys multiple informations. We analyze the relevance and importance of these features and use that information to design classifier architecture. Designing of a system for recognition of children emotions with reasonable accuracy is still a challenge specifically with reduced feature set. In this paper, Probabilistic neural network (PNN) has been designed for such task of classification. PNN has faster training ability with continuous class probability density functions. It provides better classification even with reduced feature set. LP_VQC and pH vectors are used as the features for the classifier. It has been attempted to design the PNN classifier with these features. Various emotions like angry, bore, sad and happy have been considered for this piece of work. All these emotions have been collected from children in three different languages as English, Hindi, and Odia. Result shows remarkable classification accuracy for these classes of emotions. It has been verified in standard databse EMO-DB to validate the result.