Table 3 Performance of the three models under the McAdams transform attack (UA/WA/ACC)
From: Black-box adversarial attacks through speech distortion for speech emotion recognition
αmas | CNN-LSTM (%) | GCN (%) | CNN-MAA (%) |
|---|---|---|---|
1.25 | 9.54/09.95/9.75 | 8.66/08.72/8.69 | 8.02/08.94/8.48 |
1.20 | 10.54/11.32/10.93 | 9.54/10.55/10.05 | 9.33/10.06/9.70 |
1.15 | 12.75/13.44/13.10 | 11.83/12.31/12.07 | 11.45/12.02/11.74 |
1.10 | 15.56/17.75/16.66 | 15.66/17.37/16.52 | 14.03/14.69/14.36 |
1.05 | 18.64/20.68/20.68 | 19.55/19.73/19.64 | 18.40/19.83/19.12 |
1.00 | 61.77/63.64/62.71 | 77.44/76.27/76.86 | 75.34/76.73/76.04 |
0.95 | 19.94/21.55/20.75 | 20.33/20.97/20.65 | 19.21/19.63/19.42 |
0.90 | 16.03/16.94/16.49 | 18.42/18.93/18.68 | 15.42/15.88/15.65 |
0.85 | 14.88/15.32/15.10 | 13.03/15.64/14.34 | 14.03/14.86/14.45 |
0.80 | 9.57/10.04/9.81 | 9.93/11.01/10.47 | 08.32/09.07/8.70 |
0.75 | 8.57/09.04/8.81 | 8.93/09.77/9.35 | 7.32/7.88/7.60 |