COVID-19 detection from audio: seven grains of salt

Harry Coppock, Lyn Jones, Ivan Kiskin, Björn Wolfgang Schuller

Abstract: Digital mass testing for COVID-19 via a mobile phone application may be possible through machine learning’s (ML) ability to identify patterns in data. COVID-19 appears to confer unique features in the audio produced by infected individuals and ML COVID-19 detection from breath, cough and speech audio recordings has yielded promising results. In this critique, we present seven major issues with this research and argue that further investigation is needed before conclusions about the detectability of COVID-19 from audio can be made. Many of these issues relate to a single question: are the learnt audio representations, which correlate with COVID-19 in the various collected datasets, truly audio biomarkers caused by COVID-19?