Speech Synthesis on LibriSpeech (WER, SIM, NISQA, MOSNET)
3.37WERGround Truth
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ground TruthInference Scheme=Ground Truth2026.03 | 3.37 | 91.5 | 4.62 | 4.5259 | |
| HierEASInference Scheme=HierEAS (MSpoofTTS)2026.03 | 5.32 | 90.1 | 4.602 | 4.4158 | |
| EASInference Scheme=EAS2026.03 | 5.76 | 90.2 | 4.571 | 4.3298 | |
| HierRASInference Scheme=HierRAS2026.03 | 5.91 | 90.2 | 4.596 | 4.3486 | |
| RASInference Scheme=RAS2026.03 | 6.41 | 90.5 | 4.553 | 4.2772 | |
| OriginalInference Scheme=Original (vanilla top-k sampling)2026.03 | 6.94 | 89.4 | 4.462 | 4.3418 |