Innovative Vocal Analysis Reveals Links to Type 2 Diabetes

A groundbreaking study published in the IEEE Journal of Biomedical and Health Informatics has unveiled an innovative approach to detecting Type 2 diabetes using vocal patterns. Researchers at Klick Labs have developed visual sound frequency maps, or spectrograms, analyzing voice recordings from participants to identify new diabetes-associated patterns.

Study Methodology

The research involved analyzing 14,759 voice recordings from 461 individuals, sorted by gender. Each recording consisted of short, randomly selected English sentences. These recordings were transformed into visual spectrograms, enabling an Audio Spectrogram Transformer to identify complex acoustic patterns. Notably, this approach diverges from traditional methods that rely on predefined voice characteristics, allowing for a deeper analysis of vocal biomarkers.

Findings and Accuracy Rates

The results demonstrated promising accuracy rates in distinguishing individuals with Type 2 diabetes based solely on their vocal characteristics. The models achieved a balanced accuracy of 71.3% for women and 72.4% for men, with their respective area under the receiver operating characteristic curve (AUC) values at 0.788 and 0.779.

Yan Fossat, Senior Vice President of Digital Health R&D at Klick Applied Sciences, emphasized the significance of the findings, stating, “Now, we can literally see many more clues related to diabetes within vocal biomarkers.” By utilizing relationships across various frequencies and time, this study expands on previous research and highlights combinations of features that were previously overlooked.

Building on Previous Research

This latest study builds on Klick Labs' earlier research from 2023, which focused on analyzing fixed one-sentence voice samples alongside demographic and health information, including age and body mass index (BMI). In that study, 14 specific acoustic features were identified as being linked to Type 2 diabetes.

While the current investigation reports balanced accuracy—taking into account both sensitivity and specificity—the earlier study conveyed conventional accuracy rates of 89% for women and 86% for men. This differentiation in metrics arises from variations in research design, methodology, and available data.

Expanding Research Horizons

Klick Labs has been actively broadening its exploration of vocal biomarkers beyond diabetes, recently investigating conditions such as hypertension and ovulation. Their ongoing studies are examining the following:

• The impact of audio data compression on the effectiveness of vocal biomarker extraction (scheduled for publication in JMIR Biomedical Engineering in April 2024).
• The relationship between glucose levels and voice frequency in both Type 2 diabetes patients and those with normal glucose levels (to be featured in Scientific Reports in August 2024).
• The ideal number of voice samples and optimal recording timing required to accurately predict Type 2 diabetes (upcoming in JMIR Biomedical Engineering in June 2025).

In June 2026, Klick Labs announced its collaboration with the Mayo Clinic in Florida that will further investigate the role of vocal biomarkers across various chronic diseases.

About Klick Applied Sciences

Klick Applied Sciences comprises a diverse team of professionals, including data scientists, engineers, and biological scientists, who focus on scientific research and the development of AI/ML and software tools. Their combined expertise supports various commercial efforts within the Klick Group, which also encompasses Klick Health, Klick Media Group, Klick Consulting, and Klick Ventures.