Word Decoding on Willett 2023 (test)
5.77WERDCoND-LIFT
Evaluation results
| Method | Links | |
|---|---|---|
| DCoND-LIFT2026.03 | 5.77 | |
| RNN–CTC + LMBase architecture=RNN-CTC, Post-processing/Rescoring=Language Model (LM)2026.03 | 17.4 | |
| Transformer Seq2Seq + MFCC + BART + rescoringBase architecture=Transformer Seq2Seq, Auxiliary acoustic features (MFCC)=true, Word-level multitask objective (BART)=true, Post-processing/Rescoring=true2026.03 | 19.4 | |
| Transformer Seq2Seq + MFCC + BARTBase architecture=Transformer Seq2Seq, Auxiliary acoustic features (MFCC)=true, Word-level multitask objective (BART)=true2026.03 | 25.6 | |
| Transformer Seq2Seq + BARTBase architecture=Transformer Seq2Seq, Word-level multitask objective (BART)=true2026.03 | 26 | |
| Transformer Seq2Seq + BART + Linear DTBase architecture=Transformer Seq2Seq, Word-level multitask objective (BART)=true, Day transform protocol=Linear DT2026.03 | 28.6 | |
| RNN–CTC + MFCC + BARTBase architecture=RNN-CTC, Auxiliary acoustic features (MFCC)=true, Word-level multitask objective (BART)=true2026.03 | 29 | |
| Transformer Seq2Seq + BART + No DTBase architecture=Transformer Seq2Seq, Word-level multitask objective (BART)=true, Day transform protocol=No DT2026.03 | 30.4 | |
| RNN–CTC + BARTBase architecture=RNN-CTC, Word-level multitask objective (BART)=true2026.03 | 30.9 |