Zero-shot Text-to-Speech on LibriSpeech SNR = 0dB (test-clean)
2.58UTMOSOZSpeech
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| OZSpeechTraining Dataset=LibriTTS, Fine-tuned on noisy prompts=true2025.05 | 2.58 | 6 | 23 | 0.28 | 67 | 21.37 | 54 | 0.02 | |
| F5-TTSTraining Dataset=LibriTTS2025.05 | 1.88 | 32 | 17 | — | 41 | 51.8 | 44 | 0.29 | |
| OZSpeechTraining Dataset=LibriTTS2025.05 | 1.72 | 6 | 17 | 0.22 | 45 | 46.6 | 44 | 0.08 | |
| VoiceCraftTraining Dataset=GigaSpeech2025.05 | 1.58 | 44 | 22 | 0.29 | 40 | 57.4 | 55 | 0.02 | |
| VALL-ETraining Dataset=LibriTTS2025.05 | 1.44 | 93 | 7 | 0.11 | 36 | 102.68 | 52 | 0.07 | |
| NaturalSpeech 2Training Dataset=LibriTTS2025.05 | 1.33 | 23 | 12 | 0.26 | 48 | 38.27 | 56 | 0.01 |