Speech Editing on RealEdit
4.31WERCosyEdit2
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CosyEdit2Architecture=End-to-End2026.05 | 4.31 | — | — | — | — | — | — | — | 97.91 | 3.93 | 3.21 | 0.25 | |
| CosyEditmode=one-shot in-context inference2026.01 | 4.5 | 0.9734 | 3.19 | 0.29 | 3.3 | 0.25 | 4.15 | 4.04 | — | — | — | — | |
| CosyEditArchitecture=End-to-End2026.05 | 4.5 | — | — | — | — | — | — | — | 97.34 | 4.94 | 3.19 | 0.29 | |
| SSR-Speech2026.01 | 5.05 | 0.9831 | 3.32 | 0.14 | 3.34 | 0.12 | 4.11 | 4.09 | — | — | — | — | |
| SSR-SpeechArchitecture=Cascaded2026.05 | 5.05 | — | — | — | — | — | — | — | 98.31 | — | 3.32 | 0.14 | |
| FluentSpeech2026.01 | 5.97 | 0.9274 | 2.72 | 0.78 | 2.81 | 0.67 | 2.7 | 2.6 | — | — | — | — | |
| FluentSpeechArchitecture=Cascaded2026.05 | 5.97 | — | — | — | — | — | — | — | 92.74 | — | 2.72 | 0.78 | |
| GroundTruth2026.01 | 6.06 | — | 3.34 | — | 3.38 | — | 4.21 | — | — | — | — | — | |
| GroundTruth2026.05 | 6.06 | — | — | — | — | — | — | — | — | — | 3.34 | — | |
| VoiceCraft2026.01 | 6.55 | 0.9712 | 3.18 | 0.24 | 3.31 | 0.2 | 4.04 | 4.08 | — | — | — | — | |
| VoiceCraftArchitecture=Cascaded2026.05 | 6.55 | — | — | — | — | — | — | — | 97.12 | — | 3.18 | 0.24 | |
| Ming-UniAudio2026.01 | 9.98 | 0.967 | 3.13 | 0.33 | 3.18 | 0.3 | 3.79 | 3.84 | — | — | — | — | |
| Ming-UniAudioArchitecture=End-to-End2026.05 | 9.98 | — | — | — | — | — | — | — | 96.7 | 5.36 | 3.13 | 0.33 | |
| Step-Audio-EditX2026.01 | 10.76 | 0.9588 | 3.94 | 0.61 | 3.89 | 0.54 | 3.41 | 3.49 | — | — | — | — | |
| MiMo-Audio2026.01 | 16.86 | 0.9371 | 3.48 | 0.5 | 3.38 | 0.47 | 3.55 | 3.05 | — | — | — | — |