AI-Based human audio processing for COVID-19: A comprehensive overview

Abstract

The Coronavirus (COVID-19) pandemic impelled several research efforts, from collecting COVID-19 patients' data to screening them for virus detection. Some COVID-19 symptoms are related to the functioning of the respiratory system that influences speech production; this suggests research on identifying markers of COVID-19 in speech and other human generated audio signals. In this article, we give an overview of research on human audio signals using 'Artificial Intelligence' techniques to screen, diagnose, monitor, and spread the awareness about COVID-19. This overview will be useful for developing automated systems that can help in the context of COVID-19, using non-obtrusive and easy to use bio-signals conveyed in human non-speech and speech audio productions.

Keywords: Audio processing; COVID-19; Computational paralinguistics; Digital health.

Fig. 1

Capturing and processing audio signals including speech for COVID-19 applications.

Fig. 2

Groups (given on the x-axis) that collected and analysed cough, speech, and breathing data as indicated. Although some groups collected all three types of data, they have reported their results based on the analysis of only one of them. The y-axis indicates the frequencies of the healthy and C19 subjects present in the data set. Coughvid, VoiceMed and Spira have reported number of data points; we report here number of subjects. The data sets from Cambridge, Coswara, and Coughvid are publicly available. C & B: Cough & Breath; C, S & B: Cough, Speech & Breath. On the x-axis, reference to bibliography is given in square brackets & number without bracket refers to footnote.

Fig. 3

Acoustic features’ & Machine learning techniques’ usage with the performance reported by different groups (on x-axis) for detecting COVID-19. The first row ‘+C19 subjects’ gives the C19 positive subjects’ count used by the respective groups; sequence of groups same as in Fig. 2. The features used by each group are indicated by the block colour: MFCC; SG: Spectrograms; VFO: Vocal fold Vibrations. Performance reported in the form of A: Accuracy, Se: Sensitivity, Sp: Specificity, and AUC. LR: Logistic regression. ‘Coswara’ and ‘Coughvid’ have not done any analysis with the data set they collected, hence blank blocks are shown for them. The results reported by ’Cambridge’ are: Combined analysis using cough and breath, C: Cough only and B: Breath only. On the x-axis, reference to bibliography is given in square brackets & number without bracket refers to footnote.