Analysis of cough sound observed by different cough etiquettes

Several behavioural measures have been strongly encouraged by public health services to help prevent the spread of COVID-19, including safe coughing etiquette, whereby a person coughs into their elbow to prevent water droplets spreading from person to person. These new attitudes towards coughing are likely to introduce a notable change in the audio characteristics of recorded cough sounds, causing them to be "muffled" or "dulled" as the higher frequency content is attenuated. The aim of this work is to determine if cough etiquette encouraged by public health services has altered the characteristics of a cough sound in a significant manner. To determine if a significant difference exists across the cough etiquettes observed, a statistical analysis of features commonly used by state-of-the-art cough detection algorithms is presented. The features analysed are the global means of mel-frequency cepstral coefficients, zero-crossing rate, and spectral centroid. A total of 246 individual coughs were recorded and made available for public use. Results from two-tailed t-tests (significance value of 0.05) show that a significant difference exists between means in the feature sets across the cough etiquettes observed.