An automated stress detection model based on dual approach of clinical psychologist prediction and machine learning

The timely detection of stress has become increasingly difficult because of the growing number of stressed individuals and the lack of availability of diagnostic tools. Clinical psychology mainly relies on manual detection methods, which, although effective, depend on the experience and expertise of the doctor. To address this issue, the proposed research is undertaken where a system is proposed to be developed to assess stress levels using speech samples of the target individual. This study aimed to identify stress levels from recorded audio based on acoustic characteristics. Data from 800 voice recordings of 100 participants at the Lakshmibai National Institute of Physical Education (LNIPE), North East Regional Center (NERC) Guwahati, were collected based on responses to eight questions prepared in consultation with the clinical psychiatrist. This is done so that, while answering the questions by any participant, the state of the mind of that individual is expressed. The model employs Mel Frequency Cepstral Coefficients (MFCC) and Linear Prediction Cepstral Coefficients (LPCC) with a Linear Support Vector Machine (SVM) classifier using the binary kernel, achieving an accuracy of 88%, marking a promising step towards automatic detection of stress.