New trends in non-invasive techniques for detection of hypertension through the human voice

Abstract

Early detection of hypertension based on speech processing becomes a new research field to transform health monitoring and healthcare. Such an innovative method was improved by investigating the correlation between speech properties and human physiological and emotional reactions, especially blood pressure measurements. In this study, our strategy integrates machine learning algorithms and speech feature extraction to build a blood pressure prediction model. A database of speech recordings and blood pressure values is used with the Gradient Decent algorithm and CNN architecture for training and disease classification. The achieved results presented an accuracy of $76 \%$ and a hypertension recognition rate of $78 \%$. These results could be further improved by adding PPG or ECG measurements. This work demonstrated the potential of deep speech analysis to enable early detection and diagnosis of hypertension and better prevention of cardiovascular morbidities.