A Computerized approach for emotion recognition using voice

Abstract

The purpose of this paper is providing guess information about emotion status of human using voice cut, previous machine learning researches for emotion status of human using voice cut and accuracy and recall of each research and comparison of current research and previous related researches. deeply discuss how to use machine learning for emotion status of human using voice. Firstly, explain ML using many categories Supervised learning algorithms, Unsupervised learning algorithms like that logistic regression, random forest, svm, gaussianNB, decision tree and k-nn .Python scipy provide facilities for analyze voice and provide frequency spectrum and amplitude of voice . fast Fourier transform (FFT) is providing maximum amplitude from all frequencies and frequency of maximum amplitude. The mel scale is a scale of pitches that human earshot usually observes to be intermediate from each other. Maximum mel value, maximum amplitude from all frequencies, frequency of maximum amplitude, age of user and gender are used to find out emotion status human. Part of this research is analyzing various type of voices according to emotional situation. Then train and test model using Machine learning for predict what emotional situation of is given voice. Elicited emotional speech database are creating emotional situation artificially by accumulating fact from the talker. emotional states are identified by researcher after features are trained and evaluated by machine learning models. The best recognition rate is reported by SER between 75% and 82% on random forest and recognition rate between 75% and 77% on decision tree. From this result, we can see that Decision-making algorithms (random forest and decision tree) often perform better with our data base. Therefore, we concluded that the K-Nearest Neighbor (accuracy 50%) , svm (46%), GaussianNB (46%) and logistic regression (46%). Rule based algorithms are good for recognize emotion status using voice. Random forest is best machine learning algorithm for recognize emotion status using voice according to our experiment. Random forest has good test accuracy (between 80% to 90%).