Study: Your Voice May Reveal If You Have Type 2 Diabetes

Fact checked by Nick Blackmer

Key Takeaways

• A new artificial intelligence (AI) model may be able to tell a person has type 2 diabetes with just short clips of the person’s voice.

• The researchers say that people who have diabetes can develop diabetic neuropathy or nerve damage, which can affect their vocal function, potentially leading to changes in the pitch and strength of their voice.

• Screening and diagnosis for type 2 diabetes currently involves questionnaires, blood tests, and glucose tolerance tests.

The most common and accurate way to diagnose diabetes, including prediabetes and type 2, is through blood tests. However, a new study suggests that type 2 diabetes could also be detected in a person’s voice.

Researchers from Klick Labs have created an artificial intelligence (AI) tool that can determine if a person has type 2 diabetes using just six to 10 seconds of their voice, combined with basic health information like age, sex, height, and weight. The AI model was 89% accurate in diagnosing type 2 diabetes in women and 86% accurate in men.

“Our vision is to create a screening method that is easy, convenient, and alleviates the burden and associated costs with the current blood tests,” Yan Fossat, the Vice President of Klick Labs and principal investigator of the study, told Verywell. “Voice-based screening is extremely accessible compared to the standard blood tests. A voice screening tool could be implemented outside of labs or doctors’ offices, and using people’s cell phones.”

Here’s what else you need to know about the study, including how diabetes can affect a person’s voice and whether experts think an AI model could become a new screening tool for diabetes.

Related: Type 1 vs. Type 2 Diabetes

Listening for Clues

For the study, Fossat and colleagues recruited 267 participants in India. They noted that 192 of the participants (79 women and 113 men) did not have diabetes. The other 75 (18 women and 57 men) had previously been diagnosed with diabetes.

All the participants used a smartphone app to record themselves saying a six- to 10-second fixed phrase up to six times a day for two weeks. The researchers analyzed the 18,465 recordings that were collected to listen for 14 “acoustic characteristics.” The idea was that these sounds could show vocal differences between people with type 2 diabetes and people who didn’t have the condition.

The researchers also considered other vocal features—like changes in pitch, strength, and intensity—that the human ear can’t pick up on. With help from a technique called signal processing and voice analysis software, the researchers were able to note subtle changes in the voices of people with type 2 diabetes. They used this data to train a machine learning model to be able to make predictions about type 2 diabetes based on a voice clip.

Jaycee Kaufman, a research scientist at Klick Labs and first author of the study, told Verywell that the team identified differences in vocal features between males and females with type 2 diabetes.

For example, Kaufman said voice pitch and variability of pitch were affected in women, while strength or intensity of the voice and variety of strength were affected in men.

“We believe this difference may stem from the fact that men and women experience the complications of type 2 diabetes differently, which ultimately impacts the voice differently,” said Kaufman. “Specifically, men may experience more muscle weakness associated with type 2 diabetes, whereas women may experience more edema.”

According to Kaufman, other researchers have used voice to predict neurodegenerative diseases like Alzheimer’s and Parkinson’s.

“The production of voice is a complicated process that involves the combined effects of the circulatory system, respiratory system, muscular system, nervous system, and other systems in the body,” said Kaufman. “Anything that affects these systems may have an effect on the voice, which was the motivation for this work.”

Related: How to Get Care for Type 2 Diabetes

Why Diabetes Can Affect Your Voice

People living with diabetes can develop nerve damage called diabetic neuropathy. In some cases, nerve damage leads to voice problems like bilateral vocal fold paralysis. In this condition, both vocal cords and the voice box can become partially or completely paralyzed.

Type 2 diabetes is also associated with other health problems, such as muscle weakness and swelling. According to Kaufman, previous studies have shown that the swelling can affect the voice.

“Peripheral neuropathy may damage the nerves in the larynx, resulting in hoarseness or vocal strain, and muscle weakness may be apparent in the muscle in the vocal cords or respiratory system,” said Kaufman. “In addition, the swelling associated with edema may affect the elastic and vibrational qualities of the vocal cords, which could affect the pitch.”

Giving Undiagnosed Diabetes a Voice

The study is one of the first to show how AI could help detect chronic disease, Steven Malin, PhD, FACSM, an associate professor of metabolism and endocrinology in the Department of Kinesiology and Health at Rutgers University, told Verywell. Malin was not involved in the study.

However, Malin said that without more research, it’s too early to say whether the technology will make diagnosis faster, cheaper, or more accessible for patients. While diagnosing diabetes from a voice clip is intriguing, more studies need to be done to see if the technology would work for people of different races and ages.

That said, there’s promise in having another screening tool in our arsenal.

“Even if these have ‘false-positive’ assessments, the identification of people at risk can afford people the opportunity to make behavioral changes that mitigate risk,” Malin said.

"We believe this difference may stem from the fact that men and women experience the complications of type 2 diabetes differently, which ultimately impacts the voice differently."

Jaycee Kaufman

Other experts who were not involved in the study said that the findings come at an important time, as the number of people with diabetes is rising and we need better ways to screen for it. According to the Centers for Disease Control and Prevention Diabetes Statistics Report, 14.7% of all adults age 18 or older in the United States had diabetes in 2019. And for 3.4% of the U.S. population, diabetes went undiagnosed.

“With increasing prevalence of diabetes, there is a clear need to increase screening in order to allow earlier detection of diabetes, and thus, diagnosis and treatment,” Priya Jaisinghani, MD, an endocrinologist and obesity medicine specialist at NYU Langone Health, told Verywell.

Fossat said that a non-intrusive and readily available tool like the AI-voice method they’ve come up with has the potential to screen a lot of people and could be especially useful for patients living in remote areas. More screenings could help reduce the number of people living with undiagnosed diabetes.

“We believe screening performed on a smartphone would be more accessible than a blood test and could assist in screening populations who may have more difficulties accessing healthcare,” said Fossat. “Furthermore, a mobile application for screening has the potential to reach vast amounts of the at-risk population, potentially aiding millions of people in receiving a type 2 diabetes diagnosis.”

How Is Diabetes Currently Diagnosed?

Providers can use different tests to detect and diagnose diabetes. Jaisinghani said these tests usually involve measuring blood glucose levels, which help show if someone has diabetes as well as what type they have. Tests like blood glucose (sugar), glucose tolerance, and hemoglobin A1C can all be used to find out if someone has diabetes.

Would Voice-Based Diabetes Tests Be Accurate on Their Own?

Jaisinghani said other factors could affect the accuracy of a diagnostic test that relies on voice or voice recordings, such as voice changes that stem from inflammatory and infectious disorders, neurological diseases such as myasthenia gravis, nerve injury leading to vocal cord paralysis, irritation from smoking or acid reflux, psychological or somatic conditions, and vocal stress.

“There also may be change to voice that needs to be further investigated or distinguished across age, gender, ethnicity, exposure to environmental factors, and socioeconomic demographics through further testing,” said Jaisinghani.

"We believe screening performed on a smartphone would be more accessible than a blood test and could assist in screening populations who may have more difficulties accessing healthcare."

Yan Fossat

Kaufman added that it is reasonable to consider that vocal damage could affect the test’s accuracy.

“Among our next steps, we plan to explore the effect of sickness, smoking, and vocal damage on the test’s efficacy,” she said.

Since the research was based on a limited sample of people from India, Malin said that additional studies in different countries and with a diverse group of participants would be needed, especially since diseases differ across race and ethnicity.

According to Malin, another limitation of the study is that the data came from non-smokers, so it’s not generalizable to people who smoke. The data also wouldn’t apply to people who have changes in their speech from other factors, like neurological or speech disorders.

What the Researchers Are Planning Next

Experts agree that more research with larger sample sizes that are more generalizable to different populations will need to be done before an AI voice test for diabetes could be patient-ready.

Jaisinghani said it will also be important to identify how factors like how long a person has had diabetes, as well as the complications and comorbidities of the disease, affect the voice. From there, researchers will have to determine how the factors could be useful for screening people for diabetes.

Fossat’s team said they plan to do follow-up studies to gauge the effectiveness of the technique within the next year. They want to recruit new people from different parts of the world with different demographic characteristics so they can validate and improve the current model. After the follow-up study, they want to look at how other factors like illness, smoking, and vocal damage would affect the test’s effectiveness.

Learn More: Can Type 2 Diabetes Turn Into Type 1?

Read the original article on Verywell Health.