Visual Speech Recognition on LRS2 v0.4 (test)
3.7WEROurs (raw A + V)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (raw A + V)Training Data (Hours)=LRW (157) + LRS2 (224)2021.02 | 3.7 | |
| Ours (raw A + V)Training Data (Hours)=LRS2 (224)2021.02 | 4.2 | |
| TDNNTraining Data (Hours)=LRS2 (224)2021.02 | 5.9 | |
| CTC/AttentionTraining Data (Hours)=LRW (157) + LRS2 (224)2021.02 | 7 | |
| TM-seq2seqTraining Data (Hours)=MVLRS (730) + LRS2&3^v0.4 (632)2021.02 | 8.5 | |
| Ours (V)Training Data (Hours)=LRW (157) + LRS2 (224)2021.02 | 37.9 | |
| Ours (V)Training Data (Hours)=LRS2 (224)2021.02 | 39.1 | |
| TM-seq2seqTraining Data (Hours)=MVLRS (730) + LRS2&3^v0.4 (632)2021.02 | 48.3 | |
| TDNNTraining Data (Hours)=LRS2 (224)2021.02 | 48.9 | |
| KD + CTCTraining Data (Hours)=VC2^clean (334) + LRS2&3^v0.4 (632)2021.02 | 51.3 | |
| Conv-seq2seqTraining Data (Hours)=LRW (157) + LRS2&3^v0.0 (698)2021.02 | 51.7 | |
| CTC/AttentionTraining Data (Hours)=LRW (157) + LRS2 (224)2021.02 | 63.5 | |
| LIBSTraining Data (Hours)=MVLRS (730) + LRS2 (224)2021.02 | 65.3 | |
| MV-WASTraining Data (Hours)=LRS2 (224)2021.02 | 70.4 |