Could a 20-Second Voice Recording Help Screen for Type 2 Diabetes?

Imagine screening for type 2 diabetes by reading a short story into your phone. No appointment to get started—just 20 seconds of your voice.

That’s the possibility behind research presented at the 2026 annual meeting of the European Association for the Study of Diabetes (EASD). Researchers found that artificial intelligence could identify subtle differences in speech associated with type 2 diabetes, potentially helping more people find out whether they need a blood test.

The idea is promising, but it’s still being studied. A voice recording cannot confirm a diabetes diagnosis.

What Can Your Voice Reveal?

Previous research has linked type 2 diabetes with changes in voice quality, including hoarseness, roughness and differences in breath and voice control while speaking.

Researchers at technology company thymia, working with colleagues at RMIT University in Australia, developed an AI model to recognize those patterns. They trained it using more than 63,000 voice samples from over 21,000 people in the United Kingdom and United States who reported whether they had diabetes.

To test the model, participants recorded themselves reading about 20 seconds of one of Aesop’s fables. Those familiar stories about animals and life lessons became an unexpected tool for diabetes research.

How Well Did It Work?

In an evaluation involving 7,319 UK adults, the model assigned a higher risk score to someone who reported type 2 diabetes than to someone who did not about 80% of the time when comparing the two. That doesn’t mean it correctly diagnosed 80% of all participants.

Researchers also evaluated 801 participants who completed at-home HbA1c blood tests within three months of their recordings. HbA1c measures average blood sugar over roughly two to three months.

When compared with those blood results, the model assigned the higher risk score to the person with diabetes about 75% of the time. At the screening threshold evaluated, it identified 82% of participants with type 2 diabetes.

However, it also flagged 47% of those without diabetes as positive—a substantial number who would need follow-up testing to clarify their status.

Why Researchers Are Interested

The appeal is convenience. A recording could be collected by phone or through an app, potentially reaching people who don’t attend routine health checks.

If further research supports its use, a healthcare provider might use the result to help decide who should receive blood testing. It could offer another way to get that process started.

For someone who keeps putting off a health appointment, a short recording might be an easier first step.

What Still Needs Work?

The model performed less well in Black participants and in people with heart disease, high blood pressure or obesity. Researchers suggested that limited representation and overlapping vocal changes may help explain these differences, but more testing is needed.

The findings were presented at a scientific meeting, and the researchers say clinical validation is the next step. Further studies will help determine how reliably the tool works in everyday care and across different groups of people.

For now, this research offers an intriguing possibility: your voice may someday help make diabetes screening more accessible. Blood testing would still be needed to confirm a diagnosis, and a reassuring voice result should never prevent someone from getting a blood test when they need one.

Source: EurekAlert! news release describing research by thymia and RMIT University presented at the 2026 EASD annual meeting, September 28–October 2, in Milan, Italy.