Stress Analysis Using Speech Signal

Abstract

In the area of human–computer interaction, a number of methods have been developed. The recent popular theme is “emotional intelligence”. For research, our main target objective is to observe and analyze the effect of emotions on the performance of persons while performing the tasks. In this paper, we are imposing a new approach of stress detection and classification for students during the examination period. We used mel-frequency cepstral coefficients (MFCC) for feature extraction and support vector machine (SVM) classifier for better performance. In this system, three types of corpora have been tested and classified. Support vector machine combines with the rule-based approach with energy and fundamental frequency rules. Indian dataset is created by 50 students, including male and female both. Testing of corpus proved that native area, nationality, and living place have an effect on speech frequencies. At the end of result analysis, we can see that Indians’ normal speech frequency is nearly equal to the Mongolian’s angry frequency. And as per our target view, the results show that emotions affect the performance at an average rate of 20–30%. That is, if the person is with positive emotions, then his task will achieve 20–30% better result with high speed and opposite to this person with negative emotions will move towards the failure or will get a reduced rate in his performance about the task. The accuracy of the system achieved more than 90% for depressive stress and aggressive stress. The result proved that in the examination period, the performance of students increases in excited state and decreases in a depressive state.