Can a person’s voice reveal they have type 2 diabetes? New research suggests it can. A recent study demonstrated that a brief voice recording from a phone, analyzed with artificial intelligence (AI) technology, may be an effective tool for diagnosing this common condition.
“We have shown that people with diabetes have different voice patterns when compared to similar people without diabetes,” says a coauthor of the study, Guy Fagherazzi, PhD, the director of the department of precision health at the Luxembourg Institute of Health in Belgium.
“We believe that this technology will never be accurate enough to become a diagnostic tool for type 2 diabetes that could replace a blood test,” he says. “On the other hand, we are strongly convinced that this could one day become an efficient solution to screen for diabetes and identify at-risk individuals or potential undiagnosed cases. This could significantly reduce the worldwide diabetes burden, as half of the population with diabetes ignores it.”
Scientists Analyzed Thousands of Vocal Traits
For this investigation, 607 adults — half diagnosed with diabetes and half without — were asked to provide a voice recording of themselves reading a few sentences directly from their smartphone or laptop.
Researchers noted that participants with diabetes were generally older than those without the disease, and more likely to have obesity. The average age of women with diabetes in the study was nearly 50, versus 40 for those without diabetes, and men with diabetes were about 48 on average versus 42 for those without.
The study team analyzed 25-second voice samples using two advanced techniques — one that captured up to 6,000 detailed vocal characteristics, and a second, more sophisticated deep-learning approach that focused on a refined set of about 1,000 key features.
Incorporating basic health data including age, sex, body mass index, and hypertension status, the voice-based AI algorithm correctly identified two-thirds of women with diabetes, and 7 out of 10 men.
The AI model performed even better in women age 60 or older and in people with hypertension.
“Women are usually easier to discriminate using voice when there is a health issue. We observed this in previous research on other diseases than diabetes,” says Dr. Fagherazzi. “Hypertension is also known to affect voice parameters, so we can speculate that people who have both diabetes and hypertension have an even more distinguishable voice.”
The Voice as a Source for Disease Detection
“AI and machine learning have the ability to provide new information about the relationship between voices and diabetes,” says Kevin Peterson, MD, MPH, the vice president of primary care at the American Diabetes Association.
The study noted, however, that detection through this AI technique was in 93 percent agreement with the American Diabetes Association questionnaire-based risk score, demonstrating equivalent performances between voice analysis and a widely accepted screening tool.
Why Diabetes Might Affect the Voice
Fagherazzi and his colleagues suggest that chronic high blood sugar, fatigue, acid reflux, lower pulmonary (lung) capacities, and neuropathies (conditions that affect the nerves) are some of the major drivers that could explain why people with diabetes have different voice traits than people without diabetes.
Susan Spratt, MD, a professor of medicine with a specialty in endocrinology, metabolism, and nutrition at Duke University School of Medicine in Durham, North Carolina, suspects that diabetes may affect the voice in several ways.
“First of all, diabetes can cause dehydration, which can affect vocal cord tissue, as well as the tissue lining the mouth and the tongue,” says Dr. Spratt, who was not involved in the new study. She speculates that dehydration might make words sound more staccato or “sticky.”
She adds, “Long term, diabetes can affect the nerves, including those involved with hearing. Hearing loss is more pronounced in patients with diabetes, which can also impact speech.”
“It is not one or another voice characteristic that can explain the difference of voice signatures between people with and without diabetes — it is more a combination of small changes that, when put together, can help to discriminate between the two groups,” says Fagherazzi.
“Some people with more than 10 to 20 years of diabetes experience voice changes that can be detectable by the human ear, but overall, it is mostly thanks to the progress of audio signal processing and AI that we can now detect these subtle changes,” says Fagherazzi.
Voice Diagnostics Aren’t Ready for Wide Use Yet
According to Dr. Peterson, who was not involved in the study, more research is needed before voice tests can be widely used as a tool in doctors’ offices.
“This is a hypothesis-generating study,” he says. “It points out possibilities to pursue. It is important to determine what effect such technology would have on a ‘real’ population before considering implementation. This is a new field of work that is interesting, but it is still too early to know if it will provide clinical value.”
