Lip-to-speech synthesis on LRS2-BBC (test)
4.1664UTMOSSLD-L2S
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SLD-L2SNFE=10, reference speaker embedding=true2026.02 | 4.1664 | 4.5706 | 0.8005 | 0.3954 | 0.756 | |
| NaturalL2SNFE=1, reference speaker embedding=false2026.02 | 3.5014 | 3.6702 | 0.7838 | 0.3919 | 0.562 | |
| V2SFlowNFE=30, reference speaker embedding=true2026.02 | 3.4556 | 3.9157 | 0.8064 | 0.3417 | 0.82 | |
| Ground Truth2026.02 | 3.0538 | 3.3008 | — | 0.0135 | — | |
| DiffV2SNFE=1000, reference speaker embedding=false2026.02 | 2.9575 | 3.2721 | 0.7103 | 0.4758 | 0.569 | |
| LipVoicerNFE=400, reference speaker embedding=false2026.02 | 2.3756 | 2.7973 | 0.6906 | 0.2492 | 0.578 | |
| IntelligibleNFE=1, reference speaker embedding=true2026.02 | 2.3315 | 2.3437 | 0.7932 | 0.365 | 0.701 |