Table 3.

Voice analysis related papers.

Reference	Year	Topic	Data and Cohort	Recording Device	ML Models Used	Data Processing Methods	KPIs
[71]	2020	Disease Identification: COVID-19	Private—116 subjects (76 8 weeks post COVID-19, 40 Healthy	Smartphone, Various Microphones	VGG19	Log-mel spectrogram	Acc: 0.85%, Sens: 0.89%, Spec: 0.77%
[72]	2021	Disease Identification: COVID-19	Coswara—166 subjects (83 COVID-19 positive, 83 Healthy)	Various Microphones	NB, Bayes Net, SGD, SVM, K-NN, Adaboost algorithm (model combination), DT, OneR, J48, RF, Bagging, Decision table, LWL	Fundamental Frequency (F0), Shimmer, Jitter and Harmonic to Noise Ratio, MFCC or Spectral Centroid or Roll-Off	Best overall results for vowels a, e, o: Random Forest: Acc: 82.35%, Sens: 94.12%, Spec: 70.59%
[73]	2021	Disease Identification: COVID-19	Coswara—1027 subjects (77 COVID-19 positive (54M, 23F), 950 Healthy (721M, 229F))	Various Microphones	SVM, SGD, K-NN, LWL, Adaboost and Bagging, OneR, Decision Table, DT, REPTree	ComParE_2016, FF, Jitter and Shimmer, Harmonic to Noise Ratio, MFCCs, MFCC $\Delta$ and $\Delta$$\Delta$, Spec. Centroid, Spec. Roll-off	Best overall results for vowels a, e, o: SVM: Acc: 97.07%, F1: 82.35%, Spec: 97.37%
[74]	2021	Disease Identification: COVID-19	Private—196 subjects (69 COVID-19, 130 Healthy)	Mobile App, Web App–Smartphone, Various Microphones	SVM, RBF, RF	1024 embedding feature vector from D-CNN	Best model: RF: Acc: 73%, F1: 81%
[75]	2022	Disease Identification: COPD	Corpus Gesproken Nederlands—Cohort n.s.	Various Microphones	SVM	Mean intensity (db), Mean frequency (Hz), Pitch variability (Hz), Mean center (Hz) of gravity Formants, Speaking rate, Syllables per breath group, Jitter, Jitter ppq5, Shimmer, Shimmer apq3, Shimmer apq5, HNR, ComParE_2016	Acc: 75.12%, Sens: 85%
[76]	2021	Disease Identification: COPD	Private—49 subjects (11 COPD exacerbation, 9 Stable COPD, 29 Healthy)	Smartphone	LDA, SVM	Duration, the four formants, mean gravity center, some measures of pitch and intensity, openSMILE, eGeMAPS, # of words read out loud, duration of file	p < 0.01
[77]	2021	Disease Identification: COVID-19	Coswara—Dataset 1: 1040 subjects (965 non-COVID), Dataset 2: 990 subjects (930 non-COVID)	Smartphone	LR, MLP, RF	39-dimensional MFCCs + $\Delta$ and $\Delta$$\Delta$ coeff., window size of 1024 samples, window hop size = 441 samples	Dataset 1 - RF: Average AUC: 70.69%, Dataset 2 - RF: Average AUC: 70.17%
[78]	2020	Disease Identification: COVID-19	Israeli COVID-19 collection—88 subjects (29 positive, 59 negative)	Smartphone	Transformer, SVM	Mel spectrum transformation	/z/: F1: 81%, Prec: 82%, counting: F1: 80%, Prec: 80%, /z/, /ah/: F1: 79%, Prec: 80%, /ah/: F1: 74%, Prec: 83%, cough: 58%, Prec: 72%
[79]	2021	Disease Identification: COVID-19, Asthma	COVID-19 sounds—1541 Respiratory Sounds	Mobile App, Web App–Smartphone, Various Microphones	light-weight CNN	MMFCC, EGFCC and Data De-noising Auto encoder	COVID-19/non-COVID-19 + breath + cough: Acc: 89%, Asthma/non-asthma + breath + voice Acc: 84%
[80]	2022	Disease Identification: Asthma	Private—8 subjects (100 normal, 321 Wheezing, 98 Striding, 73 Rattling sounds)	N/A	DQNN, Hybrid machine learning	IWO, Signal Selection: EHS algorithm	Spec: 99.8%, Sens: 99.2%, Acc: 100%
[81]	2022	Disease Identification: Asthma	18 patients—300 respiratory sounds, 10 types of breathing	N/A	DENN	IWO Algorithm for Asthma Detection & Forecasting	Spec: 99.8%, Sens: 99.2%, Acc: 99.91%
[82]	2020	Disease Identification: Asthma	Private—95 subjects (47 asthmatic, 48 healthy)	Various Microphones	SVM	ISCB using openSMILE, SET A: 5900 features, SET B: 6373 features, MFCC	/oU/ All feature groups: Acc: 74%
[13]	2020	Disease Identification: COVID-19	Private–240 acoustic data—60 normal, 20 COVID-19 subjects	Smartphone	LSTM (RNN)	Spec. Centroid, Spec. roll-off, ZCS, MFCC (+$\Delta$$\Delta$)	Cough: F1: 97.9% acc: 97%, breathing: F1: 98.8% acc: 98.2%, voices: F1: 92.5% acc: 88.2%
[83]	2020	Disease Identification: Asthma	88 recordings: 1957 segments (65 Severe resp. distress, 216 Asthma, 673 Mild resp. distress)	Smartphone	LIBSVM	Acoustic features: Interspeed 2010 Paralinguistic Challenge, 38 LLDs and 21 functionals	Acoustic Features: Acc: 86.3%, Sens: 85.9%, Spec: 86.9%
[84]	2021	Disease Identification: Asthma	Private—30 subjects	N/A	RDNN	Discrete Ripplet-II Transform	Proposed EAP-DL: Acc: 86.3%, Sens: 85.9%, Spec: 86.9%
[85]	2022	Symptom Identification: Voice Alteration	OPJHRC Fortis hospital in Raigarh—Cohort, not specified	Various Microphones	K-NN, SVM, LDA, LR, Linear SVM, etc.	Formant Frequencies, Pitch, Intensity, Jitter, Shimmer, Mean Autocorrelation, Harmonic to Noise ratio, Noice to Harmonic ration, MFCC, LPC	Decision Tree K-fold: Acc: 90% Sen: 90% Spec: 90%
[86]	2019	Symptom Identification: Voice Alteration	Private—Cohort n.s.	Various Microphones	Pretrained from Intel OpenVIVO and TensorFlow	Not specified, however models are vision based	N/A
[87]	2021	Disease Identification: COVID-19	Coswara, Cambridge DB-2—4352 Web App users, 2261 Android App users	Smartphone	SVM	MFCC	Acc: 85.7%, F2: 85.1%

Note. ML models: SVM = Support Vector Machine; K-NN = K-Nearest Neighbors; DT = Decision Trees; RF = Random Forest; NN = Neural Network; D-CNN = Deep Convolutional Neural Network; MLP = Multilayer Perceptron; NB = Naive Bayes; IWO = ImprovedWeed Optimization; DENN = Differential Evolutionary Neural Network; RBF model = Radial Basis Function model; LR = Linear Regression; LWL = Locally Weighted Regression (or Lowess); LDA = Linear Discriminant Analysis. Data Processing Methods: MFCCs = Mel-Frequency Cepstral Coefficients; CIF = Cochleagram Image Features; EGFCC = Enhanced-Gamma-tone Frequency Cepstral Coefficients; MMFCC = Modified Mel-frequency Cepstral Coefficients; IWO = Improved Weed Optimization; EHS = Effective Hand Strength; ISCB = Improved Standard Capon Beamforming; LPC = Linear Predictive Coding; FF = Fundamental Frequency; ZCS = Zero Crossing Rate. Metrics: Acc = Accuracy; Sens = Sensitivity; Spec = Specificity; Prec = precision; AUC = Area Under Curve.