Table 3 DSR as a function of the \({p_{fake}}\)
From: Towards the universal defense for query-based audio adversarial attacks on speech recognition system
Attack/(Sets) | DSR (%) | DSR (%) | DSR (%) | DSR (%) | DSR (%) | DSR (%) |
|---|---|---|---|---|---|---|
\(\user2{p}_{{\user2{fake}}}\)=0% | \(\user2{p}_{{\user2{fake}}}\)=10% | \(\user2{p}_{{\user2{fake}}}\)=25% | \(\user2{p}_{{\user2{fake}}}\)=40% | \(\user2{p}_{{\user2{fake}}}\)=50% | \(\user2{p}_{{\user2{fake}}}\)=60% | |
CS (Music-sets) | 98.00 | 98.00 | 98.00 | 76.20 | 76.20 | 5.44 |
DW (Music-sets) | 84.74 | 79.66 | 79.66 | 57.20 | 33.47 | 0.00 |
Average | 91.36 | 88.82 | 88.82 | 66.70 | 54.84 | 2.72 |
IRTA (Mini-Librispeech) | 84.00 | 84.00 | 84.00 | 84.00 | 84.00 | 3.00 |
DS (Mini-Librispeech) | 82.50 | 81.75 | 81.75 | 81.00 | 80.25 | 0.00 |
Average | 83.25 | 82.88 | 82.88 | 82.50 | 82.13 | 1.50 |