Scientists have created an artificial intelligence system capable of identifying signs of type 2 diabetes from 20 seconds of recorded speech. The algorithm detects subtle vocal changes associated with the medical condition that human listeners often fail to notice.
Researchers presented the findings at the annual conference of the European Association for the Study of Diabetes. The technology evaluates acoustic features in human speech, offering a potential non-invasive method for initial health screening.
Previous medical studies have shown that type 2 diabetes can affect vocal performance. Individuals diagnosed with the condition frequently exhibit increased voice hoarseness, vocal roughness, and reduced breath control while speaking. The artificial intelligence model analyzes a combination of these vocal traits to determine individual risk levels.
Acoustic analysis involves measuring fundamental frequencies, vocal jitter, shimmer, and spectral noise in audio recordings. Machine learning algorithms process these minute acoustic fluctuations to detect subtle physiological changes in vocal cord movement and respiratory muscle control caused by metabolic changes.
Type 2 diabetes is a long-term metabolic condition characterized by high levels of blood sugar. It occurs when the body becomes resistant to insulin or when the pancreas fails to produce sufficient amounts of the hormone. Unlike type 1 diabetes, an autoimmune disease typically diagnosed in childhood, type 2 diabetes is largely linked to metabolic factors and often develops gradually in adults. Over time, unmanaged diabetes can lead to severe health complications, including heart disease, nerve damage, and kidney failure.
Acoustic markers and training data
To train the voice analysis system, researchers gathered 63,283 speech recordings from more than 21,000 individuals located in the United States and the United Kingdom. The development team then evaluated the system using a new dataset of voice recordings.
During testing, study participants read aloud a passage from one of Aesop's fables for 20 seconds. Standardized passages such as Aesop's fables are widely utilized in speech analysis research because they provide consistent phonetic structures across different speakers. The passage contains a balanced range of speech sounds, allowing researchers to evaluate pitch, tone, and vocal control under identical conditions.
The European Association for the Study of Diabetes, which hosted the conference, is an international medical organization established in 1965. Based in Germany, the non-profit organization promotes scientific research into diabetes prevention and treatment, hosting annual meetings where global experts present developments in clinical care and diagnostic technology.
Testing accuracy against blood samples
In a trial involving 7,319 adult participants, the artificial intelligence model assigned a higher diabetes risk score to an individual with the condition than to a person without it in 80 percent of cases.
The research team subsequently compared the algorithm's assessments with laboratory blood tests from 801 participants. In that comparison, the system correctly identified 82 percent of people diagnosed with type 2 diabetes.
Blood testing remains the definitive medical standard for diagnosing diabetes, measuring blood sugar concentrations through fasting plasma glucose or glycated hemoglobin tests. Non-invasive screening tools are designed to identify individuals who should undergo formal laboratory testing rather than replace medical diagnosis entirely.
The system separated participants into low, medium, and high risk categories. Notably, blood test results confirmed that none of the participants categorized by the artificial intelligence into the low risk group had diabetes or prediabetes.
Potential for remote health screening
The authors of the study indicated that vocal analysis could eventually serve as a rapid primary screening method. Healthcare systems could implement the technology through telephone calls or mobile application software, enabling remote assessment without requiring an immediate trip to a medical clinic.
Digital health tools and smartphone-based screening solutions have expanded rapidly in telemedicine. Integrating voice analysis into mobile platforms could allow individuals to perform preliminary health checks from home, helping to identify undiagnosed cases in underserved regions.
The development follows earlier scientific research indicating that high consumption of red meat may increase the risk of developing type 2 diabetes. Experts believe that combining remote screening tools with public health awareness could improve early detection rates for metabolic conditions worldwide.
