Embedded Files
Legend:
Original Audio: Normal noisy input to the model, x = r∗ (y + b)
Target Audio: attacker's desired output from the model: y'
Attacked Audio: Original audio plus a perturbation: x + δ
Attacked Model Output: f (x + δ)
Demucs (Denoiser)
Demucs (Denoiser)
Original Audio
Target Audio
adversarial_speech_epoch_0.wavtarget_no_reverb__0.wavAttacked Audio
Model Output
adversarial_speech_best.wavadversarial_out_best.wavMP-SENet
MP-SENet
Original Audio
Target Audio
adversarial_speech_epoch_0.wavtarget__0.wavAttacked Audio
Model Output
adversarial_speech_best.wavadversarial_out_best.wavFull-SubNet+
Full-SubNet+
Original Audio
Target Audio
adversarial_speech_epoch_0.wavtarget__0.wavAttacked Audio
Model Output
adversarial_speech_best.wavadversarial_out_best.wavFRCRN
FRCRN
Original Audio
Target Audio
adversarial_speech_epoch_0.wavtarget__0.wavAttacked Audio
Model Output
adversarial_speech_best.wavadversarial_out_best.wavPage updated
Google Sites
Report abuse