Stress Classification based on Speech Analysis of MFCC Feature via Machine Learning

The current stress markers are mostly invasive, in which they require samples from the patients’ bodies, thus this research was conducted to find a non-invasive method to detect stress. This research emphasizes how stress detection can be done by using speech signal analysis techniques. Features from speech signals were used to capture stress together with machine learning functioning as the classifier to detect stress in a person. This research will show the advantages when using speech signal analysis techniques to detect stress compared with other stress markers. Stress detection based on speech signals was investigated, whereby speech signals were captured and analyzed in detecting stress and stress was then classified with machine learning. A phonetic feature which is the Mel-Frequency Cepstral Coefficient was extracted from the speech signals and the stress was detected with the Neural Network that were coded into a program system with Python programming language. The designed system which is the program was able to detect stress based on speech signal analysis techniques with machine learning. Therefore, psychological stress could be detected through speech signals by analyzing the count of pause and maximum amplitude, and stress was detected as stress and no stress with machine learning among International Islamic University Malaysia (IIUM) students.