Acoustic based Stress level identification using Deep Neural architecture

Abstract

Stress is defined in medicine as a physical, mental, or emotional factor that generates body or mental stress. Due to stress level humans may suffer from mental, physical illness and discomfort. Unattended stress may cause serious depression which leads to instability, bipolar disorder and suicidal intentions. Stress can be identified using Electrodermal activity sensor (EDA), Respiratory sensor, Holster unit, Electroencephalogram (EEG), Electrocardiogram (ECG), Speech Identifying stress using speech is less complicated and low cost, as separate sensors are not required. The speech features like MFCC (mel-frequency cepstral coefficients), TEO (Teager energy operator), TEO-CB, TEO-PWP can be used for detection of stress. Many of the researchers in literature used Speech Under Simulated and Actual Stress (SUSAS) database for training the machine to detect stress through speech. Some of the Machine Learning(ML) algorithms like Support VectorMachine (SVM), Hidden Markov Model (HMM), K-Nearest Neighbour (KNN), Neural Network (NN) Algorithm like Multilayer perceptron (MLP) and Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), RNN-Long Short Term Memory(RNN-LSTM) are also used for stress detection in literature .In this proposed work RNN-LSTM Attention based algorithm is to be implemented to identify stress levels like High level stress, Low level stress and Neutral level stress.