Speech Emotion Recognition Using EEMD, SVM & ANN

Abstract

Emotion recognition system from speech is one of most advanced topics in the electronic media. Emotion detection helps the security system to prevent the data from various attacks at the cyber world. A lot of research work has already been done into this contrast but the problem of accuracy is always there. This work has been done to categorize three emotions namely HAPPY, FEAR AND SAD using the EEMD, SVM and ANN algorithms. In this work, noise levels are taken so that the emotion can be identified even though if the voice signal is highly noised. The aim of this work is to check the accuracy of the EEMD algorithm with noisy signals in contrast to the emotion detection. We proceed as detecting the noise level and segmenting the signal for the further processing. There are two segments: first part is the training part in which the system is trained to identify the further proceedings. In this part, samples of each voice category are taken and their features are fetched after successful segmentation of the voice file and further on saved into the database. The second part is the testing part in which a voice sample is taken and all the required properties are fetched and matched with the saved database values. The closest match comes out as the category of the voice file.