Speech Synthesis on WordVoice 5A en (test)
3.974N-MOSMel-Recon
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mel-Recondescription=theoretical upper bound; reconstructed directly from Ground-Truth (GT) acoustic features using the FM vocoder2026.07 | 3.974 | 3.884 | 4.084 | |
| WordVoice-Controlmode=manual intervention mode using GT attributes2026.07 | 3.773 | 3.846 | 3.645 | |
| WordVoice-Freemode=adaptive prediction mode2026.07 | 3.725 | 3.869 | 3.512 | |
| CosyVoice3role=Baseline2026.07 | 3.643 | 3.838 | 3.324 |