Table 8.

COVID-19 speech classifier performance: For the Coswara (fast and normal speech) and the ComParE speech the highest AUCs were 0.893, 0.861 and 0.923 respectively and achieved by a Resnet50 trained by transfer learning in the first two cases and an SVM using with bottleneck features in the third case.

Dataset	ID	Classifier	Best Feature Hyperparameters	Best Classifier Hyperparameters (Optimised inside nested cross-validation)	Performance
					Spec	Sens	Acc	AUC	σAUC
Coswara	S1	Resnet50 + TL	Table 4	Default Resnet50 (Table 1 in Ref. [39])	90%	85%	87%	0.893	3 × 10−3
normal	S2	LSTM + TL	”	Table 4	88%	82%	85%	0.877	4 × 10−3
speech	S3	CNN + TL	”	”	88%	81%	85%	0.875	4 × 10−3
	S4	MLP + BNF	”	α3 = 0.25, α7 = 60	83%	85%	84%	0.871	8 × 10−3
	S5	SVM + BNF	”	α1 = 10−6, α4 = 105	83%	85%	84%	0.867	7 × 10−3
	S6	KNN + BNF	”	α5 = 50, α6 = 10	80%	85%	83%	0.868	6 × 10−3
	S7	LR + BNF	”	α1 = 102, α2 = 0.6, α3 = 0.4	79%	83%	81%	0.852	7 × 10−3
	S8	Resnet50 + PF	$\mathcal{M}=26,\mathcal{F}=2^{10},\mathcal{S}=120$	Default Resnet50 (Table 1 in Ref. [39])	84%	80%	82%	0.864	51 × 10−3
	S9	LSTM + PF	$\mathcal{M}=26,\mathcal{F}=2^{11},\mathcal{S}=150$	β3 = 0.1, β4 = 32, β5 = 128, β6 = 0.001, β7 = 256, β8 = 170	84%	78%	81%	0.844	51 × 10−3
	S10	CNN + PF	$\mathcal{M}=39,\mathcal{F}=2^{10},\mathcal{S}=120$	β1 = 48, β2 = 2, β3 = 0.3, β4 = 32, β7 = 256, β8 = 210	82%	78%	80%	0.832	52 × 10−3
Coswara	S11	Resnet50 + TL	Table 4	Default Resnet50 (Table 1 in Ref. [39])	84%	78%	81%	0.861	2 × 10−3
fast	S12	LSTM + TL	”	Table 4	83%	78%	81%	0.860	3 × 10−3
speech	S13	CNN + TL	”	”	82%	76%	79%	0.851	3 × 10−3
	S14	MLP + BNF	”	α3 = 0.55, α7 = 70	78%	83%	81%	0.858	7 × 10−3
	S15	SVM + BNF	”	α1 = 104, α4 = 10−2	78%	83%	81%	0.856	8 × 10−3
	S16	KNN + BNF	”	α5 = 60, α6 = 15	77%	83%	81%	0.854	8 × 10−3
	S17	LR + BNF	”	α1 = 10−3, α2 = 0.4, α3 = 0.6	77%	82%	80%	0.841	11 × 10−3
	S18	LSTM + PF	$\mathcal{M}=26,\mathcal{F}=2^{11},\mathcal{S}=120$	β3 = 0.1, β4 = 32, β5 = 128, β6 = 0.001, β7 = 256, β8 = 170	84%	80%	82%	0.856	47 × 10−3
	S19	Resnet50 + PF	$\mathcal{M}=39,\mathcal{F}=2^{10},\mathcal{S}=150$	Default Resnet50 (Table 1 in Ref. [39])	82%	78%	80%	0.822	45 × 10−3
	S20	CNN + PF	$\mathcal{M}=52,\mathcal{F}=2^{10},\mathcal{S}=100$	β1 = 48, β2 = 2, β3 = 0.3, β4 = 32, β7 = 256, β8 = 210	79%	77%	78%	0.810	41 × 10−3
ComParE	S21	Resnet50 + TL	Table 4	Default Resnet50 (Table 1 in Ref. [39])	84%	90%	87%	0.914	4 × 10−3
	S22	LSTM + TL	”	Table 4	82%	88%	85%	0.897	5 × 10−3
	S23	CNN + TL	”	”	80%	88%	84%	0.892	5 × 10−3
	S24	SVM + BNF	”	α1 = 10−1,α4 = 103	84%	88%	86%	0.923	4 × 10−3
	S25	MLP + BNF	”	α3 = 0.3, α7 = 60	80%	88%	84%	0.905	6 × 10−3
	S26	KNN + BNF	”	α5 = 20, α6 = 15	80%	86%	83%	0.891	7 × 10−3
	S27	LR + BNF	”	α1 = 102, α2 = 0.45, α3 = 0.7	81%	85%	83%	0.890	7 × 10−3
	S28	MLP + PF + SFS	$\mathcal{M}=26,\mathcal{F}=2^{11},\mathcal{S}=150$	α3 = 0.35, α7 = 70	82%	88%	85%	0.912	11 × 10−3
	S29	MLP + PF	$\mathcal{M}=26,\mathcal{F}=2^{11},\mathcal{S}=150$	α3 = 0.35, α7 = 70	81%	85%	83%	0.893	14 × 10−3
	S30	KNN + PF	$\mathcal{B}=100,\mathcal{F}=2^{10},\mathcal{S}=120$	α5 = 70, α6 = 15	80%	84%	82%	0.847	16 × 10−3
	S31	SVM + PF	$\mathcal{B}=80,\mathcal{F}=2^{11},\mathcal{S}=120$	α1 = 10−2, α4 = 10−3	79%	81%	80%	0.836	15 × 10−3
	S32	LR + PF	$\mathcal{B}=60,\mathcal{F}=2^{10},\mathcal{S}=100$	α1 = 104, α2 = 0.35, α3 = 0.65	69%	72%	71%	0.776	18 × 10−3