Scientists have created AI to detect diabetes by voice

Scientists have created an artificial intelligence (AI) system that can detect signs of type 2 diabetes in just 20 seconds of speech. The algorithm recognizes characteristic voice changes associated with the disease. The study results were presented at the annual conference of the European Association for the Study of Diabetes (EASD).

To train the system, the researchers used 63,283 voice recordings from more than 21 thousand people from the US and UK. The algorithm was then tested on new recordings in which participants read one of Aesop's fables for 20 seconds.

As it turns out, diabetes can affect the voice: people with the disease have previously been found to have increased hoarseness and roughness in their voices, as well as poorer breathing control during speech. The AI ​​analyzed a set of similar acoustic features that a person himself may not notice.

When tested on 7,319 adults, the algorithm assigned a higher risk to a person with diabetes than to a person without the disease 80 percent of the time.

The results were then compared with blood tests from 801 participants. In this case, the system correctly identified 82 percent of people with diabetes. The algorithm also categorized people into low, medium and high risk groups. None of the participants identified by the AI ​​as low risk had diabetes or prediabetes in their blood tests.

The authors believe that in the future, voice analysis could be used for rapid initial screening over the phone or through an app.