240 million people have undiagnosed diabetes and no easy way to get tested, so a health-tech team built a screening test out of talking

the problem

Hundreds of millions of adults worldwide have undiagnosed Type 2 diabetes, and the standard screening path requires a clinic visit and a blood draw most people never get around to

background

what everyone would do

The standard approach to closing a diabetes-screening gap is to make blood-test-based screening more accessible — cheaper tests, more clinics, more outreach programs to get people into a healthcare setting for a conventional blood draw, treating the access problem as one to be solved by scaling the existing testing infrastructure.

what they saw

Klick's researchers saw that the actual barrier to diagnosis wasn't the test's accuracy, it was that screening required a deliberate, costly action — a clinic visit and a blood draw — that a huge share of the undiagnosed population would never complete regardless of how cheap or available the test became. If a genuine physiological signal for diabetes existed somewhere in something people already did effortlessly and constantly, like speaking, screening could become a free byproduct of normal behavior rather than a separate task competing for time, money, and access.

the move

Klick Labs researchers recruited 267 participants in India, diagnosed as either non-diabetic or Type 2 diabetic under American Diabetes Association guidelines, and had them record a short phrase into a smartphone six times daily for two weeks, yielding more than 18,000 voice recordings. Analyzing the recordings for acoustic changes imperceptible to the human ear, the team identified 14 distinct vocal markers linked to Type 2 diabetes status, differing by gender — pitch and pitch deviation for women, amplitude fluctuations for men — and built a machine-learning model that could screen for the condition from just six to ten seconds of speech, requiring no blood draw, clinic visit, or dedicated medical device.

why it works

Because Type 2 diabetes produces measurable, if imperceptible to the human ear, changes in specific vocal characteristics, a smartphone microphone recording a few seconds of speech can capture a genuine diagnostic signal without requiring any blood sample or clinical equipment. Because the screening step piggybacks on an action people already take for free, using a phone to talk, it removes the cost, time, and access barriers that keep conventional blood-test screening from reaching the estimated hundreds of millions of people with undiagnosed diabetes worldwide.

the payoff

The underlying research, led with Mayo Clinic investigators, was published in Mayo Clinic Proceedings: Digital Health in October 2023, reporting 89 percent accuracy for women and 86 percent for men in distinguishing diabetic from non-diabetic voices. The Voice2Diabetes campaign built around the research won the Cannes Lions 2024 Innovation Grand Prix, and Klick's research team continued publishing follow-up work, including an August 2024 study in Scientific Reports establishing a measurable linear relationship between blood glucose levels and voice pitch across 505 participants, extending the underlying science toward potential non-invasive glucose monitoring.

where it breaks

The approach depends on the underlying vocal biomarkers genuinely correlating with the condition strongly enough to be clinically useful, and published accuracy figures around 86 to 89 percent, while promising for a low-cost screening tool, still fall short of a definitive diagnostic test and would need a blood-based confirmation for anyone flagged as high-risk. It also depends on voice-recording quality and consistency, background noise, microphone variation, or speech patterns unrelated to the underlying condition could all introduce error into a model trained on a specific, relatively small initial study population.

what came after

Voice2Diabetes and Klick Applied Sciences' broader vocal-biomarker research program are cited in health-technology and digital-diagnostics circles as an example of turning passive, already-occurring human behavior into a screening signal, and the underlying peer-reviewed research has continued to expand, including follow-up work on the relationship between glucose levels and voice pitch published in Scientific Reports, extending the initial screening concept toward potential non-invasive glucose monitoring applications.

references

1. [1]Voice 2 Diabetes – A Sound Innovation in Diabetes DiagnosisCampaigns of the World, 2025campaignsoftheworld.com
2. [2]New diabetes research in Scientific Reports links blood glucose levels and voiceEurekAlert! (AAAS), 2024eurekalert.org

keep it

same kind of clever

• 2009 · Airbnb (Brian Chesky, Joe Gebbia)Airbnb's New York bookings weren't stalled by its algorithm — they were stalled by bad photos, so the founders rented a camera and fixed the listings by hand, one apartment at a time.
• 2012 · Buc-ee'sBuc-ee's turned the gas station bathroom — the part every competitor treats as pure cost — into the actual reason to stop there
• 2006 · Cedars-Sinai Medical CenterCedars-Sinai got doctors to wash their hands by photographing the bacteria growing on their own palms