Harmonic-to-noise ratio as speech biomarker for fatigue: K-nearest neighbour machine learning algorithm

Area of Science:

• Speech science
• Biomedical engineering
• Machine learning

Background:

• Voice quality changes are noticeable after sleep loss.
• Circadian rhythm disruption leads to fatigue, affecting speech.
• The harmonic-to-noise ratio (HNR) is explored as a potential speech biomarker.

Purpose of the Study:

• To assess the efficacy of HNR in differentiating fatigued and normal voices after sleep deprivation.
• To investigate HNR as a speech biomarker for fatigue detection.

Main Methods:

• Acoustic samples of sustained vowel /a/ were recorded from 32 healthy young Indian males.
• One-night sleep deprivation was implemented.
• MATLAB statistical techniques and the k-nearest neighbour (KNN) machine learning algorithm were used to analyze HNR.

Main Results:

• Significant changes in voice characteristics, specifically HNR, were observed at 3 AM (p<0.05) after sleep deprivation.
• The KNN classifier successfully distinguished between normal and fatigued voice samples.
• HNR demonstrated effectiveness as a biomarker for detecting vocal alterations due to fatigue.

Conclusions:

• HNR can link sleep deprivation-induced fatigue to vocal changes.
• This method, using KNN classification, offers an additional acoustic biomarker for diagnosing fatigue.
• HNR analysis provides a valuable tool for objective fatigue assessment.