EarSteth: Cardiac Auscultation Audio Reconstruction Using Earbuds
EarSteth: Cardiac Auscultation Audio Reconstruction Using Earbuds
Abstract
Cardiac auscultation is often impractical in telehealth settings because it requires that physicians be co-located with patients in order to operate a stethoscope. We address this gap with EarSteth — a system that leverages consumer-grade active noise-cancelling earbuds to reconstruct cardiac auscultation audio signals. The system processes audio captured by the earbuds’ inner microphone with a machine learning model that reconstructs audio similar to what would be produced by a digital stethoscope during cardiac auscultation. We evaluate two existing audio super-resolution CNNs and further adapt them for heart sound reconstruction, resulting in a proposed model called EarStethNet. EarSteth models were trained using synchronous audio collected from 15 healthy adult participants with an earbud and a digital stethoscope. We found that EarStethNet was able to estimate interbeat interval with a mean absolute error of 36.6 ± 51.1 ms and was able to reconstruct cardiac auscultation audio with a mean log spectral distance of 1.22 dB.
