Text-to-Speech on LibriTTS clean (test)
0.018WERCosyVoice 2
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CosyVoice 22026.01 | 0.018 | — | 0.96 | 4.42 | 77.1 | 83 | — | |
| AudioCALMBaseline Type=Unified2026.06 | 0.02 | 4.02 | 0.668 | — | — | — | — | |
| CosyVoice 3.0Baseline Type=Modality-specific2026.06 | 0.022 | 3.96 | 0.697 | — | — | — | — | |
| Ground Truth2026.01 | 0.023 | — | — | 4.13 | 74.9 | 80.9 | — | |
| Ming-omni-TTSBaseline Type=Unified2026.06 | 0.025 | 3.82 | 0.553 | — | — | — | — | |
| UniFlow-AudioBaseline Type=Unified2026.06 | 0.032 | 3.5 | 0.57 | — | — | — | — | |
| F5-TTSBaseline Type=Modality-specific2026.06 | 0.033 | 3.85 | 0.616 | — | — | — | — | |
| KanadeToken Rate=25Hz2026.01 | 0.042 | — | 0.95 | 4.18 | 73 | 81 | — | |
| KMToken Rate=25Hz2026.01 | 0.043 | — | 0.95 | 4.05 | 74.9 | 75.9 | — | |
| KMToken Rate=12.5Hz2026.01 | 0.046 | — | 0.95 | 3.96 | 72 | 67 | — | |
| KanadeToken Rate=12.5Hz2026.01 | 0.059 | — | 0.95 | 4.13 | 77.1 | 77.9 | — | |
| X-Codec 22026.01 | 0.065 | — | 0.95 | 4.21 | 72 | 78 | — | |
| Mimi2026.01 | 0.066 | — | 0.95 | 3.48 | 74.9 | 73.9 | — | |
| BiCodec2026.01 | 0.078 | — | 0.95 | 4.12 | 73.8 | 78.9 | — | |
| UniMoE-AudioBaseline Type=Unified2026.06 | 0.078 | 3.52 | 0.361 | — | — | — | — | |
| PAST2026.01 | 0.08 | — | 0.95 | 4.14 | 74.9 | 78.4 | — | |
| StableCodec2026.01 | 0.09 | — | 0.91 | 3.78 | 71 | 66 | — | |
| ST2026.01 | 0.097 | — | 0.95 | 3.95 | 75 | 79 | — | |
| DualCodec2026.01 | 0.1 | — | 0.96 | 3.68 | 73 | 80 | — | |
| TiCodec2026.01 | 0.115 | — | 0.94 | 3.86 | 73.8 | 72.9 | — | |
| UniAudioBaseline Type=Unified2026.06 | 0.12 | 3.3 | 0.265 | — | — | — | — | |
| WavTokenizer2026.01 | 0.139 | — | 0.92 | 3.76 | 74.5 | 77 | — | |
| GroundtruthHybrid=–, Source=–, Params=–, Layers=–, Data (kh)=–, Ratio=–, Strategy=–, TFLOPs=–2026.03 | 1.85 | — | — | 4.14 | — | — | 1 | |
| CosyVoice 2Hybrid=×, Source=Public, Params=0.5B, Layers=24, Data (kh)=170, Ratio=1:0, Strategy=–, TFLOPs=1.782026.03 | 2.04 | — | — | 4.35 | — | — | 0.74 | |
| MamTraHybrid=✓, Source=Finetuned, Params=0.5B, Layers=24, Data (kh)=0.5, Ratio=1:1, Strategy=BlockBeg, TFLOPs=1.642026.03 | 2.26 | — | — | 4.35 | — | — | 0.75 | |
| MamTraHybrid=✓, Source=Finetuned, Params=0.5B, Layers=24, Data (kh)=0.5, Ratio=1:5, Strategy=WER, TFLOPs=1.542026.03 | 2.28 | — | — | 4.36 | — | — | 0.75 | |
| CosyVoicemodel_version=CosyVoice-300M-SFT2026.04 | 2.9 | — | — | — | — | — | — | |
| Ground Truthcalculation_method=whisper-large-v32026.04 | 2.9 | — | — | — | — | — | — | |
| MamTraHybrid=✓, Source=Finetuned, Params=0.5B, Layers=24, Data (kh)=0.5, Ratio=1:3, Strategy=BlockBeg, TFLOPs=1.572026.03 | 2.99 | — | — | 4.35 | — | — | 0.75 | |
| Llasa-1BHybrid=×, Source=Public, Params=1.3B, Layers=16, Data (kh)=250, Ratio=1:0, Strategy=–, TFLOPs=4.532026.03 | 3.07 | — | — | 4.3 | — | — | 0.4 | |
| CSLM-base2026.04 | 3.2 | — | — | — | — | — | — | |
| Zonos-v0.1Hybrid=✓, Source=Public, Params=1.6B, Layers=46, Data (kh)=200, Ratio=1:3, Strategy=BlockBeg, TFLOPs=7.322026.03 | 3.37 | — | — | 3.96 | — | — | 0.72 | |
| MamTraHybrid=✓, Source=Finetuned, Params=0.5B, Layers=24, Data (kh)=0.5, Ratio=1:11, Strategy=WER, TFLOPs=1.522026.03 | 4.08 | — | — | 4.37 | — | — | 0.75 | |
| CSLM-SFT2026.04 | 4.5 | — | — | — | — | — | — | |
| GLM-4-Voice2026.04 | 5.6 | — | — | — | — | — | — | |
| AnyGPT2026.04 | 27.9 | — | — | — | — | — | — | |
| SpeechGPT2026.04 | 29.1 | — | — | — | — | — | — | |
| Ground Truth2023.06 | — | 3.7 | — | — | — | — | — | |
| JETS2023.06 | — | 3.21 | — | — | — | — | — | |
| StyleTTS + HiFiGAN2023.06 | — | 3.32 | — | — | — | — | — | |
| StyleTTS 22023.06 | — | 3.87 | — | — | — | — | — | |
| VITS2023.06 | — | 3.21 | — | — | — | — | — |