The Hyfe app records sound continuously and sends files to a cloud server. These files are processed through algorithms which (a) detect cough-like sounds based on rapid changes in decibels picked up from your phone’s microphone and then (b) determine whether those sounds were coughs or not using a machine learning algorithm. The server tells the phone what the sound was, and the phone turns that information into visualizations (the chart and count of coughs). Some short (0.5 second), cough-like sounds are saved for later analysis and model improvement; longer sound files are immediately deleted after processing.
In technical terms, we use a combination of filters, transformations and neural network, a form of “deep learning”. The latter has been trained on a dataset of more than 270,000 labeled sounds, and is constantly improving. As with all models, it’s not perfect: sometimes it thinks a non-cough is a cough, and sometimes it misses a cough. But it’s getting better: we’re constantly working on improving our methods so as to better detect cough. And deep learning models are constantly improving in the realm of sound classification (Sejnowski).
What the app does
The Hyfe app is a wellness information tool. Just as other apps count calories (MyFitnessPal), workout repetitions (Count My Reps), snores (SnoreLab), and steps (Apple Health), Hyfe counts coughs.
What the app doesn’t do
Hyfe is not a disease diagnosis or treatment tool, nor is it a health care or medical service. Hyfe does not provide medical advice, nor should it be considered a substitute for any form of professional health services. The accuracy of the data collected and presented through Hyfe is not intended to match that of dedicated medical devices or scientific instruments.
If you have any health concerns at all you should speak to your doctor. In the case of a medical emergency, callemergency servicesimmediately.
