Robustness recording condition¶
w2v2-L |
hubert-L |
wavlm |
data2vec |
|
|---|---|---|---|---|
Overall Score |
0.0% passed tests (0 passed / 2 failed). |
50.0% passed tests (1 passed / 1 failed). |
50.0% passed tests (1 passed / 1 failed). |
50.0% passed tests (1 passed / 1 failed). |
Percentage Unchanged Predictions Recording Condition¶
Data |
Percentage Unchanged Predictions Recording Condition |
|||
|---|---|---|---|---|
w2v2-L |
hubert-L |
wavlm |
data2vec |
|
imda-nsc-read-speech-balanced-2.6.0-headset-boundary |
0.77 |
0.89 |
0.78 |
0.86 |
imda-nsc-read-speech-balanced-2.6.0-headset-mobile |
0.59 |
0.77 |
0.81 |
0.65 |
mean |
0.68 |
0.83 |
0.80 |
0.76 |
Visualization¶
Difference of predictions for baseline recording condition audio and different recording condition 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-L |
hubert-L |
wavlm |
data2vec |
|---|---|---|---|
|
|
|
|
|
|
|
|