Text-to-Speech on ESD English (test)
6.8WERCosyVoice-SFT-Shift
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CosyVoice-SFT-Shiftnumber of parameters=321 M, components=EmoSteer layer + fine-tuning2026.01 | 6.8 | 93.03 | 3.19 | |
| CosyVoicenumber of parameters=0 M, version=backbone2026.01 | 7.4 | 82.23 | 3.19 | |
| EmoShiftnumber of parameters=10 M, alpha=1, variant=Default2026.01 | 7.9 | 82.41 | 3.19 | |
| CosyVoice-SFTnumber of parameters=311 M, tuning=fully fine-tuned2026.01 | 8.8 | 92.05 | 3.16 | |
| EmoShiftnumber of parameters=10 M, alpha=3, variant=Best2026.01 | 11.6 | 81.5 | 3.13 |