Enhanced HMM Speech Emotion Recognition using SVM and Neural Classifier

classification has been a research area for decades. Identifying the correct emotion of the user helps in a lot of areas like crime investigation and other related research works. This particular thesis work has been done using 4 emotion categories in which SVM and Custom Neural Network has been used as a classifier. This thesis work is consisted of two main sections. The first section is called the training part and the second part is called the testing part. In the training part we have used different voice files of different categories like happy, sad, angry and aggressive to train the system according to the classified properties of the speech samples. To train the system feed forward method has been used and the database format is .mat. In the testing section we have used SVM to binarize the features of the input sample and Neural to finally classify the entire architecture. The custom neural network in this research work has been provided two categories of sample, the first sample is the training data set and the final sample is the testing data set. The finally accuracy of the plot comes out to be more than 90 %