Table A3.

Tested models for applying fusion strategy on the combined posteriors of the top models for each modality. In bold, the best models of each type in terms of the accuracy.

Fusion Model	Hyper Parameters	Accuracy with VAD	Accuracy without VAD	Accuracy withVAD FER and withouth VAD SER
Logistic Regression	$C=0.1$	71.6	68.18	70.72
	$C=1.0$	71.93	69.27	71.13
	$C=10$	72.76	69.47	71.47
	$C={10}^2$	76.62	73.50	75.68
	$C={10}^3$	77.78	76.98	79.35
	$C={10}^4$	77.33	77.33	79.63
	$C={10}^5$	77.55	77.68	79.70
	$C={10}^6$	77.55	77.78	79.63
SVM	kernel = ‘linear’; $C={10}^{-4}$	77.4	77.37	79.58
	kernel = ‘linear’; $C={10}^{-3}$	77.65	77.80	80.08
	kernel = ‘linear’; $C={10}^{-2}$	77.70	77.20	79.20
	kernel = ‘linear’; $C=0.1$	77.70	76.17	78.77
	kernel = ‘linear’; $C=1$	77.83	76.68	78.77
	kernel = ‘linear’; $C=10$	77.77	76.82	78.77
	kernel = ‘rbf’; $C=0.1$	60.03	59.62	62.47
	kernel = ‘rbf’; $C=1$	66.95	61.57	63.57
	kernel = ‘rbf’; $C=10$	70.10	64.30	66.48
	kernel = ‘rbf’; $C={10}^2$	70.10	64.33	66.57
k-NN	k = 10	60.32	59.30	60.33
	k = 20	58.95	57.15	59.07
	k = 30	57.63	57.78	59.17
	k = 40	56.78	57.55	59.01
	k = 50	55.90	57.00	58.65