In September 2024, a study by Klick Labs sparked curiosity. Published in the peer-reviewed journal IEEE Access, the study showed subtle voice properties are linked to hypertension. On social media, many envisioned a future where blood pressure could be measured during a conversation. Some mass media hinted at predicting hypertension this way. Sound far-fetched? To find out what voice research can do for people with high blood pressure now and in the future, Bloodpressure.me spoke with the study’s lead investigator, Yan Fossat, Klick Labs Senior Vice President. Get ready – this incredible journey into tomorrow starts today!
Can blood pressure be measured using voice? If you were to describe how it works to a patient, how would you do it?
We made an important discovery in determining that patients with hypertension can be identified from their voice. This is not quite the same as measuring blood pressure yet it could pave the way for that one day. Our system classifies individuals with having hypertension or not, but we do not know their blood pressure reading at the time we record their voice.
To develop our vocal biomarker model, we asked participants to read and record a predetermined phrase into their phone up to six times daily for two weeks by speaking into a proprietary mobile app Klick Labs developed. The app uses machine learning to analyze hundreds of vocal biomarkers. These unique, subtle patterns in the voice are used to identify hypertensive from non-hypertensive individuals.
In the future, when the app is released for public use, the system will be quite simple to use: People will simply have to read a predetermined phrase into their phone and the model will process the voice to produce a score, representing their risk/likelihood of having hypertension.
How are the results in your study compared to traditional blood pressure monitors?
Our model detected chronic high blood pressure with accuracies up to 84% for females and 77% for males. While traditional blood pressure monitors like cuffs have a higher degree of accuracy, it’s critical to remember that our research focuses on offering an accessible, scalable screening tool. It could fill a huge unmet need when you consider that so many people are not even aware that they have hypertension. And that’s not because blood pressure cuffs are complex or inaccurate, it is simply that most people do not have access to one. Voice-based screening could be much more accessible than the standard measurement of blood pressure that is typically done once a year during a patient’s annual physical — especially in underserved and remote areas. We hope our discovery will lead to earlier intervention in this widespread global health issue.
Does it mean that the voices of people with high blood pressure always differ from the voices of people with normal blood pressure? What is different? Can a human ear hear it?
There are properties of the voice that differ between hypertensive and non hypertensive groups, including the variability in pitch (fundamental frequency), the patterns in speech energy distribution (Mel-frequency cepstral coefficients), and the sharpness of sound changes (spectral contrast). These are very subtle mathematical properties of the voice that the human ear cannot detect.
No wonder that your research attracted so much attention. At what stage are you now? What are the plans for the future?
We are energized by our research and are exploring a number of additional conditions for vocal biomarkers. We’ve publicly released studies on glucose levels, diabetes, and hypertension and we are researching a number of other medical areas.
Do you think that in the future we can talk to our smartphone, and it will return an accurate enough blood pressure reading?
We envision a future where that is possible.. Right now our goal is to first develop a screening tool that enables people around the world to speak into their smartphones to easily and quickly find out their likelihood of having a specific condition like hypertension. While more R&D is required until we can take an actual reading of blood pressure at the moment the patient is speaking, having a screening tool is a positive first step to getting there..
Your study says that you ‘propose a machine learning-based approach that utilizes acoustic characteristics of overt speech to develop predictive models for hypertension screening.’ Is it possible to predict the onset of a chronic disease? What is possible now, using the results of your research?
We demonstrated that voice analysis can discern hypertensive subjects from not hypertensive. So, yes, we can detect hypertension from voice. What we don’t know yet is the relationship over time between hypertension and the voice. For example, if a healthy subject rapidly becomes hypertensive, would their voice change immediately or is this a cumulative effect that comes from long-term hypertension? Or, do vocal changes precede hypertension?
We know that over time, hypertension damages a number of systems in the body – from arteries, to kidneys, to eyes, etc. More research is needed to identify if voice changes are due to this cumulative damage or if they are more related to the actual blood pressure at the time of recording.
