Table 4:
J-ratios and F1-Avg scores for three input features with and without adversarial learning on DIAC-WOZ data. ‘+Adv’ denotes adversarial training
Input Features | J-Ratio | F1-Avg |
---|---|---|
Mel-spec | 4.81 | 0.619 |
Mel-spec+Adv | 4.60 | 0.646 |
Raw-audio | 2.17 | 0.646 |
Raw-audio+Adv | 1.84 | 0.660 |
Wav2vec2.0 | 1.56 | 0.686 |
Wav2vec2.0+Adv | 1.52 | 0.690 |