AI Voice App Can Detect High Blood Pressure

Klick Labs announced on Tuesday a new method for detecting chronic high blood pressure (hypertension) using voice recordings. Their research presents a non-invasive approach that could improve early hypertension detection.

In the study, 245 participants recorded their voices up to six times a day for two weeks using a Klick Labs mobile app. The app, which uses machine learning, analyzed vocal biomarkers to identify hypertension with up to 84 percent accuracy for females and 77 percent for males. Key features analyzed included pitch variability, speech energy distribution, and sound sharpness.

Yan Fossat, Senior Vice President at Klick Labs, noted, “we discovered a more accessible way to detect hypertension, which we hope will lead to earlier intervention for this widespread global health issue. Hypertension can lead to a number of complications, from heart attacks and kidney problems to dementia.”

Hypertension affects over 35% of the global population, and it is dubbed the “silent killer” by the World Health Organization, because many people are unaware they have it. Traditional blood pressure measurements, like arm cuffs, require specialized equipment and expertise, making them less accessible in some areas.

“Voice technology has the potential to exponentially transform healthcare, making it more accessible and affordable, especially for large, underserved populations,” said Jaycee Kaufman, Klick Labs research scientist and co-author of the study.

Klick Labs’ research builds on previous studies that have linked speech characteristics to heart failure symptoms and pulmonary hypertension. However, the researchers argue that this is the first study to directly investigate the relationship between speech and arterial blood pressure.

While the study’s results are promising, the researchers acknowledge several limitations. They pointed out that the number of hypertensive cases was limited, and that the participant pool was predominantly of Indian ethnicity.

They emphasized the need for a larger and more diverse participant pool to include individuals with a broader range of hypertension symptoms and varied ethnic backgrounds.

Although the proposed model demonstrated high performance with single-recording tests, the researchers observed that optimal results were achieved when data from multiple recordings per participant were used, necessitating data collection over several sessions.

They called for further research to explore methods for reducing the number of required recordings, ideally transitioning to a one-shot approach.

The researchers also noted that participants had moderate training in intonation, articulation, and speech corpus, and suggested that more naturalistic speech collection scenarios should be investigated.

As the availability and accurateness of AI-powered health tools continues to grow, it’s essential to approach them with caution. While these tools can be valuable resources, as to yet, they should not replace the advice of a qualified medical professional.