Continual Speaker Identity Unlearning on LibriHeavy and LibriSpeech After Request 3
3W-RTGU
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TGUZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 3 | 2.6 | 58.2 | 60.3 | 54.6 | 18 | |
| UNZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 3 | 2.5 | 58 | 63.8 | 55.5 | 10.6 | |
| CORTISZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 2.8 | 2.6 | 55.7 | 17.2 | 14.8 | 12.4 | |
| SGUZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 2.7 | 2.2 | 31.5 | 23.3 | 10.1 | 19.2 | |
| SelFTZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 2.7 | 2.3 | 54.8 | 55.3 | 43.4 | 11 | |
| OriginalZero-shot voice-cloning=true, Mean over 3 generation seeds=true2026.05 | 2.1 | 2.5 | 64.9 | 72.1 | 67.4 | 86.6 |