Researchers have developed an artificial intelligence tool that can analyze a short voice recording to identify people who may be at higher risk of type 2 diabetes, potentially offering a faster and non-invasive approach to screening.
The technology was developed by researchers from deep-tech company thymia and RMIT University in Melbourne, Australia. The AI model analyzes subtle changes in speech, including increased hoarseness and roughness and changes in the ability to control breathing and voice while speaking.
How the AI Tool Works
The model was trained using 63,283 voice samples from 21,129 people in the United Kingdom and the United States. Researchers then evaluated it using 20-second recordings of participants reading one of Aesop’s fables.
In one evaluation involving 7,319 adults in the UK, the model assigned higher risk scores to people who reported having type 2 diabetes than to those who did not report the condition in about 80% of cases. The researchers said this result indicates clinically useful potential, but it does not mean the tool has an 80% diagnostic accuracy.
A second evaluation involved 801 participants who underwent HbA1c blood testing within three months of providing their voice recordings. The model gave higher risk scores to people whose blood-test results indicated type 2 diabetes in 75% of cases. It also showed potential for distinguishing between low-, medium- and high-risk groups.
Blood Tests Are Still Needed
The researchers emphasize that the technology is not intended to replace conventional diabetes testing. In the study, the model correctly identified 82% of participants confirmed to have type 2 diabetes through blood testing, while its false-positive rate was 47%.
The researchers suggest that, following further clinical validation, voice analysis could potentially be used to identify people who should undergo confirmatory blood tests. Because voice samples can be collected through smartphones or telephone calls, the approach could potentially help reach people who do not regularly attend health checks.
The findings are being presented at the European Association for the Study of Diabetes annual meeting in Milan. Further clinical testing is needed to establish how well the technology performs across different populations and healthcare settings.