Breathing in Your Ear

Wearable devices are a natural choice as a means of collecting physiological data from individuals. When these wearables take the form of a watch or earphones, for example, it requires no effort on the part of the wearer outside of their normal, daily activities. While there are many metrics that can be captured, respiratory rate (RR) is particularly useful in that it can act as an indicator of one’s overall physical health.

A group of engineers at Apple have recently developed a means of transparently capturing RR with AirPods headphones. Rather than using the typical sensors — thermistors, respiratory gauge transducers, or acoustic sensors — to capture this type of data, the team used audio samples of the wearer’s breathing sounds.

Audio samples were collected from 21 individuals, aged 22 to 60, before, during, and after strenuous exercise. The samples were then manually annotated to identify inhalations and exhalations. This data was then used to train a Long-Short Term Memory network with convolutional layers. The model was taught to recognize heavy-breathing patterns, which are defined as taking more than 25 breaths per minute.

A concordance correlation coefficient of 0.76 and mean squared error of 0.2 were observed when validating the model, which shows audio to be a variable option for estimating RR. This finding implies that asthmatic episodes, heart attacks, chronic obstructive pulmonary disease, congestive heart failure, and other chronic clinical conditions may be detectable via audio collected by earphones.

This work presents a proof of concept for estimation of RR with a noninvasive wearable device. The longitudinal data that can be collected by such a device has the potential to have a real impact on health outcomes. There is no word yet as to whether or not this technology will eventually find its way into a future Apple product.