Voice Conversion on LJSpeech (target speaker)
3.22WERGround truth
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Ground truthSSL model=WavLM2025.05 | 3.22 | 5.47 | — | — | 3.85 | |
| Eta-WavLMSSL model=WavLM2025.05 | 3.81 | 5.63 | 92.46 | 47.6 | 4 | |
| Utterance stdSSL model=WavLM2025.05 | 4.13 | 7.32 | 90.34 | 51.58 | 3.8 | |
| WavLMSSL model=WavLM2025.05 | 4.56 | 5.84 | 89.52 | 52.77 | 3.84 | |
| Vector quantizationSSL model=WavLM2025.05 | 4.79 | 6.08 | 90.05 | 51.98 | 3.9 | |
| SoftSSL model=WavLM2025.05 | 4.82 | 5.94 | 91.81 | 50.11 | 3.84 | |
| PerturbationSSL model=WavLM2025.05 | 6.29 | 7.32 | 91.69 | 50.64 | 3.45 |