A rapid, non-invasive method for fatigue detection based on voice information.

Area of Science:

• Biomedical Engineering
• Human Factors Engineering
• Machine Learning Applications

Background:

• Fatigue, a result of energy depletion, impairs operator physiological and psychological states, reducing efficiency and increasing accident risk.
• Current fatigue detection methods are often subjective (e.g., scales) or require specialized equipment, limiting real-time, non-invasive monitoring.
• Speech contains acoustic biomarkers reflecting physiological and psychological status, offering a potential avenue for fatigue assessment.

Purpose of the Study:

• To develop and validate a rapid, non-invasive fatigue detection method using speech analysis.
• To correlate speech features with physiological and psychological fatigue indicators.
• To establish a machine learning model for real-time fatigue level assessment.

Main Methods:

• Constructed a fatigue model using sleep deprivation in 15 participants.
• Collected physiological data (P300, salivary glucocorticoid) and fatigue questionnaires.
• Extracted speech features and applied machine learning to correlate them with fatigue levels.

Main Results:

• Developed a fatigue detection model based on speech analysis.
• Achieved 94% accuracy in fatigue judgment using unitary voice features.
• Reached 81% accuracy using long speech features.

Conclusions:

• Speech-based acoustic analysis provides an easy, rapid, non-invasive, and efficient method for real-time fatigue detection.
• This approach can be integrated with information technology and big data for broader applications in monitoring operator fatigue.