Robustness spectral tilt

Overall scores

	CNN14	w2v2-b	hubert-b	axlstm
Overall Score	25.0% passed tests (2 passed / 6 failed).	75.0% passed tests (6 passed / 2 failed).	75.0% passed tests (6 passed / 2 failed).	50.0% passed tests (4 passed / 4 failed).

Change Ccc Downward Tilt

Threshold: -0.05

Data	Change CCC Downward Tilt
	CNN14	w2v2-b	hubert-b	axlstm
iemocap-2.3.0-emotion.dimensions.test.gold_standard	-0.05	-0.01	-0.01	-0.02
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	-0.02	-0.00	-0.00	-0.01
mean	-0.04	-0.01	-0.01	-0.01

Change Ccc Upward Tilt

Threshold: -0.05

Data	Change CCC Upward Tilt
	CNN14	w2v2-b	hubert-b	axlstm
iemocap-2.3.0-emotion.dimensions.test.gold_standard	0.10	-0.02	-0.03	0.04
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	-0.12	-0.00	-0.02	-0.08
mean	-0.01	-0.01	-0.03	-0.02

Percentage Unchanged Predictions Downward Tilt

Threshold: 0.8

Data	Percentage Unchanged Predictions Downward Tilt
	CNN14	w2v2-b	hubert-b	axlstm
iemocap-2.3.0-emotion.dimensions.test.gold_standard	0.59	0.97	0.98	0.85
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	0.47	0.99	0.99	0.72
mean	0.53	0.98	0.98	0.78

Percentage Unchanged Predictions Upward Tilt

Threshold: 0.8

Data	Percentage Unchanged Predictions Upward Tilt
	CNN14	w2v2-b	hubert-b	axlstm
iemocap-2.3.0-emotion.dimensions.test.gold_standard	0.36	0.61	0.70	0.39
msppodcast-2.6.1-emotion.dimensions.test-1.gold_standard	0.36	0.78	0.75	0.28
mean	0.36	0.70	0.72	0.34

Visualization Downward Tilt

Difference of predictions for original audio and audio with a downward spectral tilt. 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.

CNN14	w2v2-b	hubert-b	axlstm