Speech Recognition on Turkish video dataset (noisy)
18.1WER+ crosslingual pretrain + aux + kl loss
Evaluation results
| Method | Links | ||
|---|---|---|---|
| + crosslingual pretrain + aux + kl losslambda_aux=0.32020.11 | 18.1 | 5 | |
| + ce pretrain + ce loss, mid, toplambda_ce=0.62020.11 | 18.2 | 4.2 | |
| + ce pretrain + ce loss, midlambda_ce=0.62020.11 | 18.4 | 3.1 | |
| + aux + kl losslambda_aux=0.32020.11 | 18.5 | 3.1 | |
| + crosslingual pretrain2020.11 | 18.6 | 2.3 | |
| + aux losslambda_aux=0.32020.11 | 18.8 | 1.1 | |
| + kl losslambda_aux=0.32020.11 | 18.8 | 1.4 | |
| baseline2020.11 | 18.9 | — | |
| + ce pretrain2020.11 | 18.9 | 0.9 |