ASR Rescoring on MTDialogue (test)
0.0749WERMATE-MSE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MATE-MSEAlignment Loss=MSE2023.05 | 0.0749 | 0.1141 | |
| MATEAlignment Loss=Contrastive2023.05 | 0.0764 | 0.117 | |
| BERT-textFine-tuning=Full2023.05 | 0.0905 | 0.1388 | |
| MATE-NAAlignment Loss=None2023.05 | 0.0905 | 0.139 | |
| WavLM-adapterParameter-Efficient Tuning=true, Adapter location=WavLM encoder2023.05 | 0.0915 | 0.1402 | |
| ME-adapterParameter-Efficient Tuning=true, Adapter location=Masked Encoder (BERT)2023.05 | 0.0919 | 0.1412 | |
| Frozen-MEParameter-Efficient Tuning=true, Frozen Layers=Masked Encoder (BERT)2023.05 | 0.0921 | 0.1422 | |
| Multi-modal-GPT2Model Type=Uni-directional multi-modal2023.05 | 0.0924 | 0.1417 | |
| LAS rescoringHead=LAS rescoring (attention based LSTM decoder)2023.05 | 0.0927 | 0.1415 | |
| GPT2-textFine-tuning=Full2023.05 | 0.0932 | 0.1437 | |
| No rescoring2023.05 | 0.0947 | 0.1463 |