Artificial intelligence that analyses speech may offer a rapid, low-cost way to screen people for type 2 diabetes, according to a new real-world study. The approach could allow voice samples to be collected remotely by phone or via an app, potentially reaching people who miss routine health checks.
How the tool works
The speech-screening model was developed by researchers at the tech company thymia in collaboration with RMIT University in Melbourne. It was trained to recognise subtle voice changes linked to type 2 diabetes—such as greater hoarseness and reduced breath control—using more than 63,000 voice samples from over 21,000 people in the UK and the US. For testing, participants provided 20-second recordings of themselves reading Aesop's fables aloud.
Key Findings
In a UK cohort of 7,319 people, the AI model assigned higher risk scores to individuals who reported having type 2 diabetes 80% of the time. In a second analysis of a subgroup of 801 participants who completed home blood tests within three months of their recordings, the model gave higher risk scores to those with abnormal blood results 75% of the time.
"This is the largest real-world study of speech-based screening for type 2 diabetes to date which also checks the model's predictions against blood test results," said Giedre Cepukaityte, a research scientist at thymia, who will present the findings at the European Association for the Study of Diabetes (EASD) in Milan.
Limitations and Next Steps
The researchers emphasise that the tool is a screening aid, not a diagnostic test, and should not replace blood testing. The model’s performance was lower for Black patients—a result the team attributes mainly to the small number of Black participants in the training and test sets—and was also reduced for recordings from people with heart disease, high blood pressure or obesity. Such disparities highlight the need for broader, more diverse data and careful clinical validation.
Future work will evaluate the model in clinical settings and across different population groups to ensure reliability and fairness. If validated, speech screening could offer a scalable way to flag people for conventional diagnostic testing and help reach those less likely to attend in-person health checks.
Expert caution
Dr Lucy Chambers, head of research impact and communications at Diabetes UK, said AI-based tools could help identify more people who may benefit from diagnostic blood tests but warned they must be "rigorously designed and tested to make sure no-one slips through the net."
