Speech Recognition on LRS3 high-resource
17.6WER (V)USR 2.0
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| USR 2.0Model size=Huge, Pre-training dataset=LRS2+LRS3 (labelled) and VoxCeleb2+AVSpeech (unlabelled), Shared params=true2026.02 | 17.6 | 0.9 | 0.8 | |
| USR 2.0Model size=Large, Pre-training dataset=LRS3+Vox2, Shared params=true2026.02 | 21.5 | 1.3 | 1 | |
| USRModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=true2026.02 | 22.3 | 1.2 | 1.1 | |
| USR 2.0Model size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=true2026.02 | 24.8 | 1.4 | 1.2 | |
| Lip2VecModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 26 | — | — | |
| USRModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=true2026.02 | 26.5 | 1.6 | 1.3 | |
| BRAVEnModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 26.6 | 1.2 | — | |
| RAVEnModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 28.2 | 1.4 | — | |
| VATLMModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 28.4 | — | 1.2 | |
| AV-data2vecModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 28.5 | 1.3 | 1.3 | |
| AV-HuBERTModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 28.6 | 1.3 | 1.4 | |
| BRAVEnModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 28.8 | 1.4 | — | |
| u-HuBERTModel size=Large, Pre-training dataset=LRS3+Vox2, Shared params=true2026.02 | 29.1 | 1.5 | 1.3 | |
| AV-data2vecModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 32.9 | 1.7 | 1.4 | |
| RAVEnModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 33.1 | 1.9 | — | |
| Lip2VecModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 34.1 | — | — | |
| VATLMModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 34.2 | — | 1.7 | |
| AV-HuBERTModel size=Base+, Pre-training dataset=LRS3+Vox2, Shared params=false2026.02 | 34.8 | 2 | 1.8 |