Classification of stress based on speech features

Contemporary life is filled with challenges, hassles, deadlines, disappointments, and endless demands. The consequent of which might be stress. Stress has become a global
phenomenon that is been experienced in our modern daily lives. Stress might play a
significant role in psychological and/or behavioural disorders like anxiety or
depression. Hence early detection of the signs and symptoms of stress is an antidote towards reducing its harmful effects and high cost of stress management efforts. This research work thereby presented Automatic Speech Recognition (ASR) technique to stress detection as a better alternative to other approaches such as chemical analysis, skin conductance, electrocardiograms that are obtrusive, intrusive, and also costly. Two set of voice data was recorded from ten Arabs students at Universiti Utara Malaysia (UUM) in neural and stressed mode. Speech features of fundamental, frequency (f0); formants (F1, F2, and F3), energy and Mel-Frequency Cepstral Coefficients (MFCC) were extracted and classified by K-nearest neighbour, Linear Discriminant Analysis and Artificial Neural Network. Result from average value of fundamental frequency
reveals that stress is highly correlated with increase in fundamental frequency value. Of
the three classifiers, K-nearest neighbor (KNN) performance is best followed by linear
discriminant analysis (LDA) while artificial neural network (ANN) shows the least performance. Stress level classification into low, medium and high was done based of the classification result of KNN. This research shows the viability of ASR as better means of stress detection and classification.