Watermark Robustness on VCTK, LibriSpeech, and LJSpeech (test)
100Accuracy (ACC)TimbreWM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TimbreWMSpeech Reconstruction Method=Vocos2026.06 | 100 | 0 | |
| AlignMarkSpeech Reconstruction Method=Vocos2026.06 | 100 | 0 | |
| VoiceMarkSpeech Reconstruction Method=Vocos2026.06 | 99 | 4 | |
| AlignMarkSpeech Reconstruction Method=EnCodec2026.06 | 99 | 2 | |
| AlignMarkSpeech Reconstruction Method=HiFiGAN†2026.06 | 99 | 1 | |
| VoiceMarkSpeech Reconstruction Method=EnCodec2026.06 | 96 | 16 | |
| AlignMarkSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 96 | 11 | |
| VoiceMarkSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 95 | 21 | |
| VoiceMarkSpeech Reconstruction Method=HiFiGAN†2026.06 | 94 | 22 | |
| AlignMarkSpeech Reconstruction Method=FACodec†2026.06 | 93 | 24 | |
| TimbreWMSpeech Reconstruction Method=HiFiGAN†2026.06 | 92 | 48 | |
| AlignMarkSpeech Reconstruction Method=ClearerVoice†2026.06 | 92 | 27 | |
| VoiceMarkSpeech Reconstruction Method=FACodec†2026.06 | 88 | 45 | |
| TimbreWMSpeech Reconstruction Method=ClearerVoice†2026.06 | 78 | 65 | |
| WMCodecSpeech Reconstruction Method=HiFiGAN†2026.06 | 77 | 79 | |
| VoiceMarkSpeech Reconstruction Method=ClearerVoice†2026.06 | 76 | 60 | |
| WMCodecSpeech Reconstruction Method=ClearerVoice†2026.06 | 74 | 83 | |
| AudioSealSpeech Reconstruction Method=ClearerVoice†2026.06 | 67 | 79 | |
| TimbreWMSpeech Reconstruction Method=EnCodec2026.06 | 57 | 99 | |
| TimbreWMSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 57 | 99 | |
| TimbreWMSpeech Reconstruction Method=FACodec†2026.06 | 53 | 100 | |
| WMCodecSpeech Reconstruction Method=Vocos2026.06 | 53 | 100 | |
| WavMarkSpeech Reconstruction Method=ClearerVoice†2026.06 | 51 | 100 | |
| WavMarkSpeech Reconstruction Method=EnCodec2026.06 | 51 | 100 | |
| WavMarkSpeech Reconstruction Method=FACodec†2026.06 | 51 | 100 | |
| WavMarkSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 51 | 100 | |
| WavMarkSpeech Reconstruction Method=Vocos2026.06 | 51 | 100 | |
| WavMarkSpeech Reconstruction Method=HiFiGAN†2026.06 | 51 | 100 | |
| WMCodecSpeech Reconstruction Method=EnCodec2026.06 | 51 | 100 | |
| WMCodecSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 51 | 100 | |
| AudioSealSpeech Reconstruction Method=EnCodec2026.06 | 50 | 100 | |
| AudioSealSpeech Reconstruction Method=FACodec†2026.06 | 50 | 99 | |
| AudioSealSpeech Reconstruction Method=SpeechTokenizer†2026.06 | 50 | 99 | |
| AudioSealSpeech Reconstruction Method=Vocos2026.06 | 50 | 100 | |
| AudioSealSpeech Reconstruction Method=HiFiGAN†2026.06 | 50 | 100 | |
| WMCodecSpeech Reconstruction Method=FACodec†2026.06 | 50 | 100 |