AI model uses 20-sec voice recordings to identify type 2 diabetes

Researchers developed an AI model that analyses 20-second voice recording to screen for type 2 diabetes. In a study, the model, trained on 63,283 voice samples from 21,129 people in UK and US, identified reported diabetes with higher risk scores 80% of the time. In subgroup with HbA1c tests, it correctly picked up 82% of individuals with type 2 diabetes.