Text-to-Speech on LibriSpeech-PC Subset B (test-clean)
1.95WERCosyVoice3
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CosyVoice3Type=AR, Params.=0.5B, Training Data (hours)=1000K Multi.2026.07 | 1.95 | 69 | 4.28 | |
| DELTA-TTSType=Ours, Params.=0.5B+94M, Training Data (hours)=0.585K LibriTTS2026.07 | 2.17 | 69 | 4.27 | |
| Ground-truthType=-2026.07 | 2.23 | 69 | 4.1 | |
| DiTARType=AR, Params.=0.6B, Training Data (hours)=100K Emilia2026.07 | 2.39 | 67 | 4.22 | |
| F5-TTSType=NAR, Params.=0.3B, Training Data (hours)=100K Emilia, NFE=322026.07 | 2.42 | 66 | 3.88 | |
| FireRedTTSType=AR, Params.=0.6B, Training Data (hours)=248K Multi.2026.07 | 2.69 | 47 | — | |
| MaskGCTType=NAR, Params.=1.1B, Training Data (hours)=100K Emilia, NFE=502026.07 | 2.72 | 69 | 3.9 | |
| E2 TTSType=NAR, Params.=0.3B, Training Data (hours)=100K Emilia, NFE=322026.07 | 2.95 | 69 | 3.56 | |
| CosyVoiceType=AR, Params.=0.3B, Training Data (hours)=170K Multi.2026.07 | 3.59 | 66 | 4.14 |