Subjective Speech Synthesis Evaluation on Seed-TTS and CosyVoice zh 3 (test)
0CMOSMI-DFM (GibbsTTS)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MI-DFM (GibbsTTS)Scheduler=Numerical KO2026.05 | 0 | 4.28 | |
| MI-DFMScheduler=Grid-searched2026.05 | -0.153 | 4.21 | |
| Masked DFMScheduler=Closed-form KO2026.05 | -0.224 | 4.11 | |
| Masked DFMScheduler=DiFlow-TTS2026.05 | -0.247 | 4.1 | |
| MI-DFM w/o correctorScheduler=Numerical KO2026.05 | -0.459 | 4.03 | |
| MaskGCT (original)Scheduler=MaskGCT2026.05 | -0.612 | 4.19 | |
| Masked DDScheduler=MaskGCT2026.05 | -0.647 | 3.87 |