Blood pressure monitoring from naturally recorded speech sounds: advancements and future prospects

Abstract

The development of an accurate, cuffless system for continuous monitoring of blood pressure is essential to reduce the number of deaths due to hypertension. In this study, we present a groundbreaking artificial intelligence-based system developed for accurate blood pressure prediction from spoken sentences in natural everyday situations, using only a smartphone without additional measurements. Our method uses hyperparameter-tuned machine learning (ML) techniques, including Synthetic Minority Over-sampling Technique (SMOTE), to classify blood pressure as normal or high. By automatically detecting vowels in recorded speech sentences, we extract a statistical features vector with demographic information (1 × 59-D). Experimental results highlight impressive classification accuracies, reaching 98.45% for systolic BP and 99.61% for diastolic BP with the Adaptive synthetic sampling approach for imbalanced learning (ADASYN). These findings underscore the meaningful physiological information embedded in human speech and demonstrate the potential of our hyperparameter-tuned ML methods in revolutionizing health monitoring practices, particularly in the domain of telehealth, internet of things devices and remote monitoring.