Zero-shot Text-to-Speech on Expressive Speech
10.91WERGround Truth
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ground Truth2025.08 | 10.91 | 68.7 | 1.47 | 0.87 | |
| CosyVoice 2#Hz=25, Data (#hours)=167K, #Param=581M, PT?=✓2025.08 | 11.2 | 70.6 | 0.1 | -0.16 | |
| Vevo2#Hz=12.5, Data (#hours)=101K, 7K, #Param=872M, PT?=✓2025.08 | 11.48 | 68.9 | 0 | 0 | |
| F5-TTSData (#hours)=101K, #Param=336M, PT?=✗2025.08 | 11.77 | 69.5 | -1.05 | -0.06 | |
| MaskGCT#Hz=50, Data (#hours)=101K, #Param=1048M, PT?=✗2025.08 | 13.42 | 73.6 | -1.14 | 0.07 | |
| Vevo2-base#Hz=12.5, Data (#hours)=101K, 7K, #Param=872M, PT?=✗2025.08 | 14.32 | 68.1 | -0.49 | -0.03 | |
| Vevo2-base#Hz=12.5, Data (#hours)=101K, #Param=872M, PT?=✗2025.08 | 15.52 | 67.7 | -0.84 | -0.06 | |
| Vevo2-base#Hz=12.5, Data (#hours)=7K, #Param=872M, PT?=✗2025.08 | 37.57 | 61.1 | -1.25 | -0.57 |