Voice frequency patterns reveal clues linked to type 2 diabetes

A new research study in theIEEE Journal of Biomedical and Health Informatics reveals how scientists at Klick Labs created visual maps of sound frequencies (spectrograms) to identify complex, new patterns associated with Type 2 diabetes, using only smartphone recordings of randomly selected short English sentences and participant sex as study criteria. 

The scientists analyzed 14,759 voice recordings from 461 adults, which were converted into a visual spectrogram before being analyzed by an Audio Spectrogram Transformer. The process allows the model to learn complex combinations of acoustic patterns directly from the recordings rather than relying primarily on predefined voice features selected in advance. Using voice only, the sex-stratified models achieved a balanced accuracy of 71.3 percent for women and 72.4 percent for men, with area under the receiver operating characteristic curve (AUC) values of 0.788 (78.8 percent) and 0.779 (77.9 percent), respectively.

Now, we can literally see many more clues related to diabetes within vocal biomarkers. By capturing relationships across frequencies and time, we expanded our analysis beyond shimmer, jitter, and other predefined voice features and found complex combinations of acoustic patterns and features that weren't on anyone's radar. We are energized by the findings as we continue to explore associations between vocal biomarkers and Type 2 diabetes."

Yan Fossat, Senior Vice President, Digital Health R&D at Klick Applied Sciences and principal investigator of the study

Advancing the 2023 study

The research builds on Klick Labs' 2023 Voice to Diabetes study in Mayo Clinic Proceedings: Digital Health, which analyzed fixed one-sentence smartphone voice recordings alongside participants' age, body mass index (BMI), and other basic health information. The study identified 14 predefined acoustic features, including pitch, intensity, shimmer, and jitter, associated with Type 2 diabetes. 

It should be noted that the new study reported balanced accuracy (which averages sensitivity and specificity, giving equal weight to identifying people with and without Type 2 diabetes), while the 2023 research reported conventional accuracy (the percentage of all predictions that are correct) of 89 percent for women and 86 percent for men, so the figures should not be compared directly. The difference in reported figures reflects the studies' different inputs, methods, validation approaches, and datasets, not a claim that the new models are more accurate.

Over the last few years, Klick Labs has expanded its vocal biomarker research to areas, including hypertension and ovulation, while progressing its studies within diabetes exploring:

• The impact of audio data compression on feature extraction for vocal biomarker detection (JMIR Biomedical Engineering, April 2024)

• The effects of glucose levels on voice fundamental frequency in Type 2 diabetes and individuals with normoglycemia (Scientific Reports, August 2024)

• The optimal number of voice samples and ideal recording schedule for the prediction of Type 2 Diabetes (JMIR Biomedical Engineering, June 2025)

In June 2026, Klick Labs announced it has embarked on a series of studies with Mayo Clinic in Florida, exploring the use of novel vocal biomarkers in connection with a number of chronic diseases and serious health conditions.