Detection of COVID-19 from voice, cough and breathing patterns: Dataset and preliminary results

Abstract

COVID-19 heavily affects breathing and voice and causes symptoms that make patients’ voices distinctive, creating recognizable audio signatures. Initial studies have already suggested the potential of using voice as a screening solution. In this article we present a dataset of voice, cough and breathing audio recordings collected from individuals infected by SARS-CoV-2 virus, as well as non-infected subjects via large scale crowdsourced campaign. We describe preliminary results for detection of COVID-19 from cough patterns using standard acoustic features sets, wavelet scattering features and deep audio embeddings extracted from low-level feature representations (VGGish and OpenL3). Our models achieve accuracy of 88.52%, sensitivity of 88.75% and specificity of 90.87%, confirming the applicability of audio signatures to identify COVID-19 symptoms. We furthermore provide an in-depth analysis of the most informative acoustic features and try to elucidate the mechanisms that alter the acoustic characteristics of coughs of people with COVID-19.

Deep learning and machine learning-based voice analysis for the detection of COVID-19: A proposal and comparison of architectures

“… [25] Vowel, speech, cough Unique device Unspecified None (76 recovered, 116 total) R, H CNN (transfer learning) Cross-validation 74% (mean, R vs H) Despotovic et al. [24] Vowel, speech, breath, cough Crowdsourced None (self-reported) 84 (1103 total) P, H Adaboost, Multilayer Perceptron, CNN Cross-validation 88% Muguli et al [26] – DiCOVA challenge Vowel, speech, breath, cough Crowdsourced None (self-reported) 60 (990 total) P, H Various Various 73% (baseline) 87% (best) Abbreviations: PCR: Polymerase Chain Reaction-based molecular swab; P: COVID-19 Positive subjects; H: Healthy subjects; R: Recovered subjects; CNN: Convolutional Neural Network; SVM: Support Vector Machine; RNN: Recurrent Neural Network. “Lossless” refers to raw, unprocessed and uncompressed sound data, while “lossy” implies that compression and/or artifacts are present.…”

“…[21] obtained an accuracy higher than 90% through principal component analysis (PCA) and data processing using mel-frequency cepstral coefficients (MFCC). The algorithms presented by Pinkas [22] and Shimon [23] yielded average accuracies around 83%, while Despotovic [24] achieved 88% in accuracy considering vocal, speech, cough and breathing crowdsourced sounds. To the best of our knowledge, the only study involving recovered subjects is the one by Suppakitjanusant et al.…”

Role of Artificial Intelligence in COVID-19 Detection

“…Nonetheless, this modality is not useful when the pleura is spared from the pneumonic pathology during the early course of the disease [ 297 ]. Recent developments in the diagnosis of COVID-19 using signals such as respiratory sounds, speech signals, and coughing sounds, have also attracted many researchers [ 298 , 299 ]. Furthermore, in the future, this can be combined with other imaging modalities and signals to enhance the performance of the system using various deep learning approaches.…”

A comprehensive survey on the biomedical signal processing methods for the detection of COVID-19

“…Furthermore, this study investigating the impacts of the disease through voice, analysing the voice before and after infection. According to this study [ 18 ], the detection of COVID-19 through voice by pre-screening method which leads to automatic identification of COVID-19 using analysis of TFR (time frequency representations) with same performance [ 19 ]. presenting dataset for cough, voice, audio recording for breathing gathered form individuals infected by SARS-CoV-2 virus, and also the non-infected subjects as large scale crowd sourced operation.…”