A study suggests that artificial intelligence could identify signals linked to type 2 diabetes in just 20 seconds of speech. This is still an experimental method, but it shows promise, as many patients are unaware they have the condition. The goal is not to replace traditional blood tests, but to more easily detect people who should undergo screening, especially those who rarely visit a doctor. Type 2 diabetes can develop over a long period without symptoms obvious enough to prompt a visit to a healthcare professional. This is the main challenge addressed by a new approach developed by the organization Thymia in collaboration with RMIT University in Melbourne. The principle is straightforward: Record your voice for about 20 seconds, either through a telephone or an application, and let artificial intelligence analyze characteristics that are not detectable by the human ear. The algorithm examines the tone, rhythm of speech, specific acoustic parameters, and how a person controls their breath. To train the model, researchers used 63,283 voice recordings from 21,129 people in the United Kingdom and the United States. They then tested it on 7,319 British adults, who were asked to read a passage from the Fables of La Fontaine for about 20 seconds. Among them, 217 individuals declared having type 2 diabetes. The results, presented at the European Association for the Study of Diabetes (EASD) in Milan, are encouraging. The AI assigned a higher risk score to people with diabetes in 80% of cases. Its sensitivity reached 82%, although the false positive rate was 47%.