Robustness low quality phone

Overall scores

	w2v2-b	w2v2-L	w2v2-L-robust	w2v2-L-xls-r	w2v2-L-vox
Overall Score	100.0% passed tests (4 passed / 0 failed).	25.0% passed tests (1 passed / 3 failed).	100.0% passed tests (4 passed / 0 failed).	0.0% passed tests (0 passed / 4 failed).	75.0% passed tests (3 passed / 1 failed).

Change Ccc Low Quality Phone

Threshold: -0.05

Data	Change CCC Low Quality Phone
	w2v2-b	w2v2-L	w2v2-L-robust	w2v2-L-xls-r	w2v2-L-vox
iemocap-2.3.0-emotion.dimensions.test.gold_standard	-0.03	-0.06	-0.03	-0.07	-0.04
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	-0.03	-0.02	-0.04	-0.06	-0.05
mean	-0.03	-0.04	-0.04	-0.07	-0.04

Percentage Unchanged Predictions Low Quality Phone

Threshold: 0.5

Data	Percentage Unchanged Predictions Low Quality Phone
	w2v2-b	w2v2-L	w2v2-L-robust	w2v2-L-xls-r	w2v2-L-vox
iemocap-2.3.0-emotion.dimensions.test.gold_standard	0.58	0.41	0.72	0.39	0.63
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	0.59	0.38	0.72	0.36	0.62
mean	0.58	0.40	0.72	0.38	0.62

Visualization

Difference of predictions for original audio and low quality phone audio. The allowed prediction difference \(\delta < 0.05\) is highlighted in green in the upper plot. The lower plot shows the distributions of the two predictions.

w2v2-b	w2v2-L	w2v2-L-robust	w2v2-L-xls-r	w2v2-L-vox