Continuation Zero-shot Speech Synthesis on LibriSpeech (test-clean)
1.98WERHMELLE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MELLETraining Data Hours=50K2024.07 | 1.98 | 1.47 | 0.508 | |
| VoiceboxTraining Data Hours=60K2024.07 | 2 | — | 0.593 | |
| MELLE-R2Training Data Hours=50K, reduction factor=22024.07 | 2.02 | 1.45 | 0.489 | |
| MELLE-R3Training Data Hours=50K, reduction factor=32024.07 | 2.1 | 1.52 | 0.462 | |
| MELLE-R4Training Data Hours=50K, reduction factor=42024.07 | 2.1 | 1.59 | 0.437 | |
| Ground Truth2024.07 | 2.15 | 1.61 | 0.668 | |
| MELLE-limitedTraining Data Hours=9602024.07 | 2.22 | 1.53 | 0.48 | |
| Ground Truth (mel-spectrogram)2024.07 | 2.24 | 1.64 | 0.617 | |
| MELLE-R5Training Data Hours=50K, reduction factor=52024.07 | 2.25 | 1.66 | 0.41 | |
| VALL-E RTraining Data Hours=9602024.07 | 2.32 | 1.58 | 0.363 | |
| VALL-E 2Training Data Hours=50K2024.07 | 2.32 | 1.6 | 0.504 | |
| Ground Truth (EnCodec, 8 codebooks)2024.07 | 2.33 | 1.65 | 0.593 | |
| CLaM-TTSTraining Data Hours=55K2024.07 | 2.36 | — | 0.477 | |
| ELLA-VTraining Data Hours=9602024.07 | 2.91 | 2.1 | 0.303 | |
| VALL-ETraining Data Hours=60K2024.07 | 3.8 | — | 0.508 |