Robustness simulated recording condition¶
w2v2-L |
hubert-L |
wavlm |
data2vec |
|
|---|---|---|---|---|
Overall Score |
33.3% passed tests (2 passed / 4 failed). |
16.7% passed tests (1 passed / 5 failed). |
33.3% passed tests (2 passed / 4 failed). |
16.7% passed tests (1 passed / 5 failed). |
Percentage Unchanged Predictions Simulated Position¶
Data |
Percentage Unchanged Predictions Simulated Position |
|||
|---|---|---|---|---|
w2v2-L |
hubert-L |
wavlm |
data2vec |
|
emovo-1.2.1-emotion.test |
0.59 |
0.70 |
0.79 |
0.62 |
imda-nsc-read-speech-balanced-2.6.0-headset |
0.81 |
0.82 |
0.90 |
0.80 |
timit-1.4.1-files |
0.86 |
0.77 |
0.92 |
0.84 |
mean |
0.75 |
0.76 |
0.87 |
0.75 |
Percentage Unchanged Predictions Simulated Room¶
Data |
Percentage Unchanged Predictions Simulated Room |
|||
|---|---|---|---|---|
w2v2-L |
hubert-L |
wavlm |
data2vec |
|
emovo-1.2.1-emotion.test |
0.35 |
0.61 |
0.59 |
0.48 |
imda-nsc-read-speech-balanced-2.6.0-headset |
0.44 |
0.72 |
0.66 |
0.65 |
timit-1.4.1-files |
0.55 |
0.72 |
0.64 |
0.70 |
mean |
0.45 |
0.68 |
0.63 |
0.61 |
Visualization Simulated Position¶
Difference of predictions for audio with a baseline simulated position and audio with a different simulated position. 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 |
|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|