Table 4 Performance of the three models under the Modulation Spectrum Smoothing attack(UA/WA/ACC)
From: Black-box adversarial attacks through speech distortion for speech emotion recognition
αms | CNN-LSTM (%) | GCN (%) | CNN-MAA (%) |
|---|---|---|---|
0.30 | 13.74/13.49/13.62 | 11.53/11.95/11.74 | 11.14/11.59/11.37 |
0.25 | 14.94/14.99/14.97 | 12.46/13.55/13.01 | 12.22/13.32/12.77 |
0.20 | 18.44/19.29/18.87 | 14.93/16.32/15.63 | 15.74/17.22/16.48 |
0.15 | 20.34/20.77/20.56 | 17.44/18.30/17.87 | 19.32/20.43/19.88 |
0.10 | 23.21/23.89/23.55 | 19.92/21.49/20.71 | 20.11/21.34/20.73 |
0.05 | 26.56/28.44/27.50 | 23.93/24.60/24.27 | 22.94/24.05/23.50 |