Automatic analysis of heart sounds using speech recognition techniques

Listening to the sound of a patient’s heart has long been a quick, noninvasive, and revealing way of gathering information on the state of the heart. As a complement to other instruments for measuring heart function (such as the EKG), it could be useful to have an automatic device that essentially duplicates the listening performed by a physician with a stethoscope. Many of the challenges in developing such a device – such as source variability and background noise – are directly equivalent to problems that have been dealt with in the development of automatic speech recognition. Here we propose a project to investigate how best to adapt techniques from speech recognition to the automatic monitoring and diagnosis of heart sounds.