AI tool can detect type 2 diabetes from a 20-second voice recording with 80% accuracy

From the Newshound archive · first published 29 September 2026 · last updated 30 September 2026.

Researchers at tech firm Thymia and RMIT University in Melbourne have developed an AI system that identifies likely type 2 diabetes cases by detecting hoarseness and loss of breath control in a brief voice clip. The tool was built using more than 63,000 recordings drawn from over 21,000 people in Britain and the US, and tested by having participants read passages from Aesop's fables aloud. In trials involving 7,319 UK participants, it correctly assigned elevated risk scores to people with type 2 diabetes 80% of the time. Scientists say it could allow patients to screen themselves via a phone call or app, without an immediate blood test.

Thymia researcher Giedre Cepukaityte stressed the tool 'opens a new route to screening' but is 'not a replacement for a blood test'.

Accuracy fell for Black patients, who were underrepresented in the study, and for those with conditions including heart disease, high blood pressure or obesity.

In a group of 801 participants whose home blood tests were completed within three months of their recording, the AI returned correct high-risk scores 75% of the time.