Audio-Visual Stress Classification Using Cascaded RNN-LSTM Networks.

Area of Science:

• Psychology
• Computer Science
• Artificial Intelligence

Background:

• Stress significantly impacts mental and physical health, often described as a "hidden pandemic."
• Traditional stress detection methods like self-report questionnaires (e.g., Perceived Stress Scale) and interviews have limitations in validity, specificity, and resource consumption.
• There is a need for objective, efficient, and accurate methods for stress detection to improve mental health and overall well-being.

Purpose of the Study:

• To introduce and evaluate a novel method for detecting human mental stress using audio-visual data processing.
• To highlight the importance of mental health and contribute to human well-being through advanced stress detection.
• To explore the efficacy of audio-visual stress identification as an alternative to conventional methods.

Main Methods:

• A novel approach for human mental stress detection was developed by processing audio-visual data.
• The cascaded Recurrent Neural Network-Long Short-Term Memory (RNN-LSTM) strategy was employed for data analysis.
• The method was validated on the RAVDESS dataset, which includes audio-visual recordings of various emotions.

Main Results:

• The proposed audio-visual stress detection method achieved an accuracy of 91% on the RAVDESS dataset.
• The system successfully classified eight distinct emotions and differentiated between stressed and unstressed states.
• The cascaded RNN-LSTM strategy proved effective in processing complex audio-visual cues for stress identification.

Conclusions:

• Audio-visual data processing offers a promising and accurate avenue for detecting human mental stress.
• The developed cascaded RNN-LSTM model demonstrates high efficacy in identifying stress, outperforming traditional subjective measures.
• This research contributes a valuable tool for mental health monitoring and promoting overall human well-being.