Emotional States Recognition by Speech Features, Its Need and Impact of Various ANN Model on Recognition Rate

This paper focuses on enhancing requirements of the current voice command and speaking technologies with consideration of emotional states, so that the interaction with computers and smart devices may become more human. It indicates the study of recognition of human emotional states and shows the basic idea that how signal processing can apply for the same. Experiment is divided into two major and important parts; one is the analysis of speech samples where second is associated with modeling of neural network and its training method. Further we extend our experiment to study the analysis procedure. Here we actually show how the speech features can extracted from speech samples and how feed forward back propagation algorithm used to carry the training procedure. An emotion recognition rate of approximately 70% was obtained for five basic emotional states. The final phase of experiment illustrates that repetitive training of ANN significantly affect the recognition rate as concluding decision and suggests the boundary for optimum ANN training.