Voice analysis may potentially screen and monitor diabetes

Key findings are:

• The differences between voice recordings of nondiabetic and T2DM men and women were significant (both in the entire dataset and an age-matched and body mass index-matched sample).
• The highest predictive accuracy was achieved by pitch, pitch SD, relative average perturbation jitter for women, and intensity and 11-point amplitude perturbation quotient shimmer for men.
• The optimal prediction models, incorporating age and BMI, achieved accuracies of 0.75±0.22 for women and 0.70±0.10 for men through 5-fold cross-validation in the age-matched and BMI-matched samples.

They said, “We found distinct differences between the voices of individuals with and without T2DM.”

An ensemble model with T2DM prevalence and BMI for men and women achieved maximum test accuracy of 0.89 for women and 0.86 for men. The optimal models were a 2-vocal-feature NB implementation for men and a 3-vocal-feature LR for women.

Women with T2DM had a slightly lower pitch with less variation, while men with T2DM had slightly weaker voices with more variation due to differences in disease symptom manifestations between the sexes.

Although further research with larger and more diverse cohorts is needed, our findings suggest that voice analysis could be an accessible and cost-effective screening tool for T2DM. The author writes that voice assessment could aid in early intervention and management, potentially reducing the disease burden and improving health outcomes.

Reference:

Kaufman, J., Thommandram, A., & Fossat, Y. (2023). Acoustic analysis and prediction of Type 2 diabetes mellitus using Smartphone-Recorded voice segments. Mayo Clinic Proceedings Digital Health, 1(4), 534–544. https://doi.org/10.1016/j.mcpdig.2023.08.005