Zero-shot Text-to-Speech on LibriSpeech clean (test)
1.3WERSpeechEdit
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SpeechEditParameters=0.5 B, Training Data (h)=0.8 k2026.01 | 1.3 | 0.48 | 4 | |
| SpeechEdit-Ablation-TaskParameters=0.5 B, Training Data (h)=0.8 k2026.01 | 1.5 | 0.53 | 4.02 | |
| Step-Audio-EditXParameters=3 B, Training Data (h)=-2026.01 | 1.6 | 0.63 | 3.32 | |
| SpeechEdit-Ablation-DataParameters=0.5 B, Training Data (h)=0.8 k2026.01 | 1.9 | 0.45 | 4.01 | |
| VALL-E-A1Parameters=0.5 B, Training Data (h)=5 k2026.01 | 2.1 | 0.61 | 4 | |
| VALL-E-A2Parameters=0.5 B, Training Data (h)=1 k2026.01 | 2.7 | 0.48 | 4.02 |