Whisper-to-normal conversion on AISHELL6 Whisper (test)
3.141DNSMOS Overall ScoreGround Truth
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ground Truth2025.11 | 3.141 | 3.552 | 3.795 | 3.68 | 2.868 | 3.129 | 3.395 | — | 1 | 1 | 1 | 1 | |
| WhisperVCproposed_method=true2025.11 | 3.072 | 3.506 | 3.72 | 3.562 | 2.831 | 3.352 | 3.164 | 16.932 | 0.816 | 0.675 | 0.945 | 0.785 | |
| Seed-VCmode=zero-shot2025.11 | 2.868 | 3.255 | 4.031 | 3.74 | 2.467 | 2.814 | 3.088 | 46.423 | 0.851 | 0.772 | 0.952 | 0.758 | |
| Coarse-onlyablation=True2025.11 | 2.716 | 3.338 | 3.343 | 3.258 | 2.797 | 3.186 | 2.418 | 18.729 | 0.529 | 0.638 | 0.943 | 0.761 | |
| WhisperVC + OT-CFM (Residual)ablation=True2025.11 | 2.652 | 3.253 | 3.313 | 3.425 | 2.795 | 3.139 | 2.83 | 18.266 | 0.528 | 0.65 | 0.944 | 0.76 | |
| WhisperVC + OT-CFM (Full)ablation=True2025.11 | 2.548 | 3.178 | 3.188 | 3.398 | 2.17 | 2.716 | 2.935 | 19.576 | 0.515 | 0.613 | 0.93 | 0.744 | |
| OT-CFM (Residual) w/o VAEablation=True2025.11 | 1.83 | 3.218 | 1.632 | 2.755 | 1.729 | 1.519 | 2.233 | 40.155 | 0.511 | 0.557 | 0.898 | 0.712 | |
| Whispered input2025.11 | 1.102 | 1.155 | 1.195 | 2.703 | 1.308 | -0.067 | 1.9 | 22.937 | 0.582 | 0.498 | 0.784 | 0.673 |