Automatic Speech Recognition on LibriSpeech (test)
1.97WER (%)Clean
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CleanPrefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 1.97 | — | — | |
| CleanPrefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 1.98 | — | — | |
| CleanPrefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.01 | — | — | |
| CleanPrefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.01 | — | — | |
| CleanPrefix (s)=4.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.01 | — | — | |
| CleanPrefix (s)=1.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.02 | — | — | |
| CleanPrefix (s)=2.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.02 | — | — | |
| CleanPrefix (s)=3.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.02 | — | — | |
| CleanPrefix (s)=2.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.03 | — | — | |
| CleanPrefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.03 | — | — | |
| CleanPrefix (s)=1.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.05 | — | — | |
| CleanPrefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.05 | — | — | |
| AO*Prefix (s)=2.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.16 | — | — | |
| AO*Prefix (s)=3.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 2.35 | — | — | |
| AO*Prefix (s)=2.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 3.19 | — | — | |
| AO*Prefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 4.27 | — | — | |
| AO*Prefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 4.38 | — | — | |
| Clean (No Attack)Defense=None (Raw Model)2026.06 | 4.75 | — | — | |
| Clean (No Attack)Defense=PVP Vote, Defense Type=Input Preprocessing2026.06 | 4.8 | — | — | |
| AO*Prefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 4.84 | — | — | |
| AO*Prefix (s)=1.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 5.34 | — | — | |
| Clean (No Attack)Defense=DT-AT, Defense Type=Adversarial Training2026.06 | 5.7 | — | — | |
| Clean (No Attack)Defense=CE-AT, Defense Type=Adversarial Training2026.06 | 5.72 | — | — | |
| AO*Prefix (s)=4.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 5.8 | — | — | |
| Clean (No Attack)Defense=TI, Defense Type=Adversarial Training2026.06 | 5.81 | — | — | |
| Clean (No Attack)Defense=DMW, Defense Type=Adversarial Training2026.06 | 5.86 | — | — | |
| Muting WhisperDefense=DMW, Defense Type=Adversarial Training2026.06 | 5.94 | — | — | |
| Muting WhisperDefense=TI, Defense Type=Adversarial Training2026.06 | 6.04 | — | — | |
| Muting WhisperDefense=DT-AT, Defense Type=Adversarial Training2026.06 | 6.09 | — | — | |
| Muting WhisperDefense=CE-AT, Defense Type=Adversarial Training2026.06 | 6.11 | — | — | |
| Clean (No Attack)Defense=AudioPure, Defense Type=Input Preprocessing2026.06 | 6.17 | — | — | |
| Clean (No Attack)Defense=LPF, Defense Type=Input Preprocessing2026.06 | 6.45 | — | — | |
| Clean (No Attack)Defense=WaveGuard, Defense Type=Input Preprocessing2026.06 | 6.76 | — | — | |
| Muting WhisperDefense=AudioPure, Defense Type=Input Preprocessing2026.06 | 6.8 | — | — | |
| Muting WhisperDefense=LPF, Defense Type=Input Preprocessing2026.06 | 7.52 | — | — | |
| Muting WhisperDefense=WaveGuard, Defense Type=Input Preprocessing2026.06 | 7.62 | — | — | |
| PGDDefense=DT-AT, Defense Type=Adversarial Training2026.06 | 8.13 | — | — | |
| AO*Prefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 8.28 | — | — | |
| PGDDefense=AudioPure, Defense Type=Input Preprocessing2026.06 | 8.38 | — | — | |
| AO*Prefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 8.45 | — | — | |
| PGDDefense=CE-AT, Defense Type=Adversarial Training2026.06 | 8.69 | — | — | |
| SlothDefense=AudioPure, Defense Type=Input Preprocessing2026.06 | 8.7 | — | — | |
| SlothDefense=DT-AT, Defense Type=Adversarial Training2026.06 | 8.71 | — | — | |
| AO*Prefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 8.78 | — | — | |
| SlothDefense=DMW, Defense Type=Adversarial Training2026.06 | 9.26 | — | — | |
| SlothDefense=CE-AT, Defense Type=Adversarial Training2026.06 | 9.38 | — | — | |
| SlothDefense=TI, Defense Type=Adversarial Training2026.06 | 9.38 | — | — | |
| PGDDefense=DMW, Defense Type=Adversarial Training2026.06 | 9.61 | — | — | |
| PGDDefense=TI, Defense Type=Adversarial Training2026.06 | 9.8 | — | — | |
| UniversalPrefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 9.99 | — | — | |
| UniversalPrefix (s)=4.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 10.05 | — | — | |
| UniversalPrefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 10.46 | — | — | |
| UniversalPrefix (s)=2.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 10.62 | — | — | |
| AOPrefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 11.11 | — | — | |
| UniversalPrefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 11.15 | — | — | |
| SGPrefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 11.32 | — | — | |
| PGDPrefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 11.56 | — | — | |
| UniversalPrefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 11.89 | — | — | |
| PGDPrefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.02 | — | — | |
| PGDPrefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.13 | — | — | |
| PGDPrefix (s)=4.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.18 | — | — | |
| AO*Prefix (s)=1.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.23 | — | — | |
| GT (Bound)Prefix (s)=4.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.44 | — | — | |
| AOPrefix (s)=2.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.58 | — | — | |
| PGDPrefix (s)=3.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.7 | — | — | |
| GT (Bound)Prefix (s)=3.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 12.94 | — | — | |
| UniversalPrefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 13.14 | — | — | |
| PGDPrefix (s)=2.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 13.17 | — | — | |
| PGDPrefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 13.21 | — | — | |
| AOPrefix (s)=2.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 13.42 | — | — | |
| PGDPrefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 13.75 | — | — | |
| AOPrefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.17 | — | — | |
| GT (Bound)Prefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.27 | — | — | |
| UniversalPrefix (s)=2.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.45 | — | — | |
| PGDPrefix (s)=2.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.46 | — | — | |
| GT (Bound)Prefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.46 | — | — | |
| GT (Bound)Prefix (s)=2.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.51 | — | — | |
| AOPrefix (s)=1.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.52 | — | — | |
| PGDPrefix (s)=1.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.58 | — | — | |
| AOPrefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.82 | — | — | |
| SlothDefense=LPF, Defense Type=Input Preprocessing2026.06 | 14.88 | — | — | |
| UniversalPrefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 14.91 | — | — | |
| SlothDefense=WaveGuard, Defense Type=Input Preprocessing2026.06 | 14.98 | — | — | |
| AOPrefix (s)=4.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 15.03 | — | — | |
| AOPrefix (s)=1.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 15.15 | — | — | |
| PGDDefense=WaveGuard, Defense Type=Input Preprocessing2026.06 | 15.47 | — | — | |
| PGDPrefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 15.64 | — | — | |
| GT (Bound)Prefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 16 | — | — | |
| MI-FGSMDefense=CE-AT, Defense Type=Adversarial Training2026.06 | 16.2 | — | — | |
| SGPrefix (s)=1.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 16.26 | — | — | |
| PGDDefense=LPF, Defense Type=Input Preprocessing2026.06 | 16.28 | — | — | |
| GT (Bound)Prefix (s)=3.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 16.47 | — | — | |
| MI-FGSMDefense=TI, Defense Type=Adversarial Training2026.06 | 16.85 | — | — | |
| MI-FGSMDefense=DMW, Defense Type=Adversarial Training2026.06 | 16.86 | — | — | |
| MI-FGSMDefense=DT-AT, Defense Type=Adversarial Training2026.06 | 16.98 | — | — | |
| PGDPrefix (s)=1.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 17.01 | — | — | |
| UniversalPrefix (s)=1.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 17.21 | — | — | |
| UniversalPrefix (s)=1.0, Delay (s)=0.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 17.37 | — | — | |
| AOPrefix (s)=4.0, Delay (s)=1.0, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 17.68 | — | — | |
| UniversalPrefix (s)=3.0, Delay (s)=0.5, SNR constraint=20 dB, Target window=3.0 s, Victim model=Wav2Vec 2.02026.06 | 18.52 | — | — |