Acoustic Signatures of COVID-19 in Speech, Cough and Breathing: Associations with Smoking Habits

At the end 2019, the global community witnessed the emergence of a new and severe acute respiratory disease named Coronavirus Disease 2019 (COVID-19). The gold standard diagnostic method is the reverse transcriptase polymerase chain reaction (RT-PCR) test, but it has limitations such as high costs, the need for specialized laboratories and trained personnel, and reporting times that can vary from several hours to days. In this study, we propose voice signals as digital biomarkers and therefore as an alternative tool for detecting COVID-19 positive and negative subjects. The study was conducted on 246 subjects, each of whom underwent RT-PCR testing and recorded cough, breathing, and speaking signals. We extracted Mel Frequency Cepstral Coefficients (MFCCs) from each voice signal. The results demonstrate that MFCCSs extracted from speech, cough, and respiratory signals contain significant acoustic information for distinguishing between positive and negative subjects. However, some MFCCs showed greater potential with non-smoking subjects, suggesting possible similarities in the changes caused by COVID-19 and smoking, which make diagnosis difficult.