Speech Recognition on CHiME-4 real (test)
2.24WERUSTC-iFlytek
Evaluation Results
| Method | Links | |
|---|---|---|
| USTC-iFlyteknumber of systems=52018.03 | 2.24 | |
| Proposed System (BLSTM-GEV + TDNN-LF-MMI + LSTMLM)number of systems=12018.03 | 2.74 | |
| Oracle Ocp2024.02 | 2.8 | |
| RWTH/UPB/FORTHnumber of systems=52018.03 | 2.91 | |
| MERLnumber of systems=62018.03 | 2.98 | |
| Whisper (real label)Adaptation strategy=Finetuning, Supervision type=Supervised2024.05 | 5.2 | |
| STARAdaptation strategy=STAR, Supervision type=Unsupervised2024.05 | 6 | |
| UADFfull_name=Uncertainty-Aware Dynamic Fusion, reproduced_in_paper=true2024.02 | 6.08 | |
| TOKreweight (A_t)Adaptation strategy=Self-training, Token-level re-weighting=A_t, Supervision type=Unsupervised2024.05 | 6.2 | |
| UTTfilterAdaptation strategy=Self-training, Filtering=Utterance-level, Supervision type=Unsupervised2024.05 | 6.4 | |
| TOKreweight (G_t)Adaptation strategy=Self-training, Token-level re-weighting=G_t, Supervision type=Unsupervised2024.05 | 6.5 | |
| GERreproduced_in_paper=true2024.02 | 6.75 | |
| Whisper (frozen)Adaptation strategy=None, Supervision type=Zero-shot2024.05 | 6.8 | |
| Whisper (self-train.)Adaptation strategy=Self-training, Supervision type=Unsupervised2024.05 | 6.9 | |
| Oracle Onb2024.02 | 9.1 | |
| LM rank2024.02 | 11 | |
| 1-best2024.02 | 11.1 | |
| CHiME-4 baselinenumber of systems=12018.03 | 11.51 | |
| IRTFpost-filter=Wiener2019.05 | 12.61 | |
| IRTFpost-filter=none2019.05 | 13.34 | |
| MVDRpost-filter=Wiener2019.05 | 14.05 | |
| MVDRpost-filter=none2019.05 | 14.72 | |
| GEVpost-filter=BAN2019.05 | 15.71 | |
| GEVpost-filter=none2019.05 | 19.52 | |
| Unprocessedpost-filter=none2019.05 | 19.9 |