AI voice analysis may help detect type 2 diabetes within seconds
A new artificial intelligence-based technology could help identify people at risk of type 2 diabetes through a 20-second voice recording, potentially offering a fast and non-invasive screening tool. Researchers at technology company Themia, in collaboration with Australia’s RMIT University, developed a model designed to identify voice characteristics associated with type 2 diabetes, including changes in hoarseness, roughness and breathing control during speech.
The model was trained on more than 63,000 voice recordings from about 21,000 people in Britain and the United States before being tested on short recordings of participants reading stories. In an initial test involving 7,319 people in Britain, the system assigned higher risk scores to participants who reported having type 2 diabetes in 80 percent of cases.
A second analysis of 801 people who underwent home blood testing within three months of recording their voices produced a 75 percent rate for participants with diabetes. Researchers said the model performed across different age and gender groups, although its effectiveness declined among some groups, particularly people of African descent, who were underrepresented in the training sample.
Performance also fell among people with heart disease, high blood pressure or obesity, conditions that can produce voice changes similar to those associated with diabetes. Unlike conventional diagnosis, which relies on blood testing, the technology could potentially allow preliminary screening through a smartphone or app, researchers said.
However, the researchers stressed that the system cannot replace blood tests or provide a confirmed diagnosis. Further testing in clinical settings and across more diverse populations is needed before its effectiveness can be established. The findings suggest that voice analysis could eventually serve as a rapid screening method to identify people who may require conventional testing for type 2 diabetes.
