TABLE 2. Comparison of General Cough Detection Approaches.

Ref	Disease	Method	Dataset	Size of the data	Specificity	Sensitivity	F1-score	Accuracy
[48]	Cough and non-cough (e.g., asthma, bronchiectasis, or chronic obstructive pulmonary disease)	Mel frequency cepstral coefficients (MFCC)/moment theory	Cough sound	–	90%	90%	–	–
[69]	Cough and non-cough	Artificial neural network (ANN)	Cough sound	19,832 Sound	91%	86%	88%	91%
[2]	Cough and non-cough	Logistic regression	Cough sound	1980 Sounds	99.42%	90.31%	88.74%	–
[39]	Cough and non-cough	Deep neural networks (DNN), and hidden Markov model (HMM)	Cough sound	45000 Sounds	88.6%	90.1%	88.6%	–
[29]	Cough and non-cough	Principal Component Analysis (PCA), and Deep Neural Networks (DNN)	Cough sound	810 Events	–	–	–	99.91%
[42]	Cough and non-cough	HMMs	Cough sound	–	–	–	–	82%
[34]	Cough and non-cough	SVM	Cough sound	13 cough	88.58%	92.71%	–	90.69%
[88]	Cough and non-cough	Hu moments	Cough sound	–	98.64%	88.94%	–	–
[8]	Cough and non-cough	Convolutional Neural Networks (CNNs)
[77]	Cough and snoring	K-nearest neighbor (k-NN).	Cough sound	26 Subjects	–	–	88%	–
[58]	Pertussis	MFCC	Questionnaire, Cough sound	414 Coughs	90%	92.38%	–	–
[46]	Pulmonary disease or asthma	Cough embeddings Cosine – cough detection	Cough sound	5380 Cough samples	96.37%	86.55%	–	91.46%
[18]	Pulmonary disease	Random Forest with 1000 trees (RF_1000), Adaboost, and Gradient-Boosted Tree (GBT), root-mean-square energy cough detection	Cough sound	8,491 cough samples	84.14%	74.62%	79.47%	94.6%
[14]	Tuberculosis	HMM and MFCC	Cough sound	746 Coughs	72%	95%	–	78%
[37]	Tuberculosis	SVM, DNN, sequential minimal optimization (SMO)	Cough sound	13,429 Cough frames and 43,925 non-cough frames	99.6%	75.5%	–	–
[76]	Tuberculosis	DNN, MLP, SVM	Cough sound	13,429 Cough frames and 43,925 non-cough frames	–	–	–	88.2%
[87]	COVID-19	Recurrent Neural Network (RNN) and the Long Short-Term Memory (LSTM)	Cough sound	–	–	–	97.9%	97%