Investigation and ordinal modelling of vocal features for stress detection in speech

This paper investigates a robust and effective automatic stress detection model based on human vocal features. Our study experimental dataset contains the voices of 58 Greek-speaking participants (24 male, 34 female, 26.9±4.8 years old), both in neutral and stressed conditions. We extracted a total of 76 speech-derived features after extensive study of the relevant literature. We investigated and selected the most robust features using automatic feature selection methods, comparing multiple feature ranking methods (such as RFE, mRMR, stepwise fit) to assess their pattern across gender & experimental phase factors. Then, classification was performed both for the entire dataset, and then for each experimental task, for both genders combined and separately. The performance was evaluated using 10-fold cross-validation on the speakers. Our analysis achieved a best classification accuracy of 84.8% using linear SVM for the social exposure phase and 74.5% for the mental tasks phase using the gaussian SVM classifier. The ordinal modelling improved significantly our results, yielding a best on-subject basis 10-fold cross-validation classification accuracy of 95.0% for social exposure using gaussian SVM and 85.9% for mental tasks using the gaussian SVM. From our analysis, specific vocal features were identified as being robust and relevant to stress along with parameters to construct the stress model. However, it is was observed the susceptibility of speech to bias and masking and thus the need for universal speech markers for stress detection.