Signs of type 2 diabetes can be detected in just a few seconds from the way someone talks, new research being presented at the annual meeting of the European Association for the Study of Diabetes (EASD) in Milan, Italy (Sept 28 – Oct 2) shows.
The study, the largest of its kind, concludes that AI-based analysis of short recordings could be a fast, non-invasive and scalable screening tool for the condition.
Type 2 diabetes is becoming increasingly common and early detection is key to preventing complications such as heart disease and nerve damage. Yet many cases go undetected, putting pressure on healthcare systems. In the UK, for example, about 30% of cases are undiagnosed and around 60% of the £10.7 billion the National Health Service (NHS) spends on diabetes each year goes on managing complications.
Screening currently involves blood tests or GP appointments.
For example, the NHS includes diabetes screening in the health checks it offers to people aged 40-plus every five years but fewer than half of eligible adults (40.4%) go for these useful, but time-consuming, appointments, say the study’s authors.
Previous research has linked type 2 diabetes to changes in speech such as increased hoarseness and roughness and inability to control breath and voice as well while speaking.
Researchers at deep tech company thymia, led by Roseline Polle, Senior Machine Learning Researcher, and Dr Elisa Brann, Senior Research Scientist, with colleagues at RMIT University in Melbourne, Australia, first developed an AI model to detect these changes, training it on 63,283 voice samples from 21,129 people in the UK and US who’d reported whether they’d been diagnosed with diabetes.
Then, they validated, or tested, the speech model on remote 20-second recordings by people reading one of Aesop’s fables (short stories with animal characters and moral lessons).
The first of two evaluations compared the model’s performance on recordings made by 7,319 adults (67% female, 45.8% 40 years or older , 217 self-reported having type 2 diabetes) in the UK.
This found that the speech model gave a higher risk score to individuals who reported having type 2 diabetes than to those who did not report having the condition 80% of the time, considered to be clinically useful.
The model performed well across different sexes and ages. However, performance was lower on recordings from Black participants. This was likely due to the low number of Black participants reporting type 2 diabetes, say the authors.
Performance was also lower on recordings from people with heart disease, high blood pressure or obesity. This is thought to be because these conditions often coincide with type 2 diabetes and may cause similar vocal changes.
The second evaluation involved a sub-group of 801 participants who took HbA1c tests at home within three months of the speech recording. (The HbA1c blood test measures average blood sugar levels in the past two to three months and is the gold-standard test for type 2 diabetes. It can also be used to detect prediabetes, where blood sugar levels are higher than normal but not high enough to be classed as diabetes.)
Here, the speech model gave a higher risk score to people with type 2 diabetes than to those who did not have the condition, based on HbA1c tests, 75% of the time. The sensitivity was 82% (i.e. the model correctly picked up 82% of the individuals with type 2 diabetes) and the false positive rate was 47%.
The speech model was also able to distinguish between people at low-, medium- and high-risk of type 2 diabetes, based on their HbA1c results. None of the people the model classed as low risk had blood results in the diabetic or prediabetic range.
The researchers conclude that type 2 diabetes can be detected at clinically useful levels from 20 seconds of speech.
Pending further clinical validation of the tool, one option would be for GPs to use short recordings to triage patients. Those at higher risk could then be given blood tests to confirm their status. Speech-based screening would sit alongside blood testing, not replace it, say the authors.
Giedrė Čepukaitytė, Research Scientist at thymia, who will be presenting the findings at EASD, says: “This is the largest real-world study of speech-based screening for type 2 diabetes to date which also checks the model's predictions against blood test results as well as against what people reported about their own diagnosis. Those flagged up as higher risk by the model had blood results to match.
“This has the potential to change what screening looks like. A speech sample can be taken over the phone or through an app, so we can reach far more of the people who need a blood test than current pathways do, particularly those who never get to a health check.
“Our model opens a new route to screening for diabetes. It is not a replacement for a blood test, and it should never stop anyone who thinks they need one from getting one.
“Our next step is to test the model in clinical settings and to understand how well it works for every group of people, because a screening tool has to work for everyone.”
28-Sep-2026
Emilia Molimpakis and Stefano Goria are co-founders of thymia Ltd. Elisa Brann, Roseline Polle, Giedrė Čepukaitytė, Owen Parsons, and Alexandra Livia Georgescu are employees of thymia Ltd. Emilia Molimpakis, Stefano Goria, Owen Parsons, and Alexandra Livia Georgescu hold equity in the company, which may benefit from commercialisation of technologies similar to those described in this paper.