Project Achoo: A Practical Model and Application for COVID-19 Detection From Recordings of Breath, Voice, and Cough

The COVID-19 pandemic created significant interest and demand for infection detection and monitoring solutions. In this paper, we propose a machine learning method to quickly detect COVID-19 using audio recordings made on consumer devices. The approach combines signal processing and noise removal methods with an ensemble of fine-tuned deep learning networks and enables COVID detection on coughs. We have also developed and deployed a mobile application that uses a symptoms checker together with voice, breath, and cough signals to detect COVID-19 infection. The application showed robust performance on both openly sourced datasets and the noisy data collected during beta testing by the end users.