BLOOD PRESSURE ESTIMATION FROM VOICE SPECTRUM WITH CONVOLUTIONAL NEURAL NETWORKS

*Motoki Sakai Ph.D.

ABSTRACT

Blood pressure (BP) is an important vital index for predicting the risk of brain or cardiac infarction. This study proposes a method for estimating BP from voice signals without a BP gauge. The voice can be recorded using common tools, such as smartphones, which contributes to the ease of BP monitoring. In our experiment, systolic BP (SBP), diastolic BP (DBP), and voice data were obtained from 14 male and six female subjects. Five convolutional neural network (CNN) structures were used to estimate BPs. Consequently, no CNNs could estimate SBP and DBP in both men and women. However, there were effective CNN structures for male SBP, female SBP, and female DBP estimation, which could estimate BPs with a mean absolute error (MAE) of less than 10 mmHg.

Keywords: Blood pressure, Blood pressure estimation, Voice, Convolutional neural network, preventive healthcare.