Automatic Speech Recognition on OGI-S (Age 4-7)
17.32WERSingle Dataset Fine-tune
Evaluation Results
| Method | Links | |
|---|---|---|
| Single Dataset Fine-tuneFine-tuning Protocol=Encoder+Projector+LoRA2026.06 | 17.32 | |
| C-DR MoERouting Strategy=Soft Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine), Entropy-Aware Routing (EAR)=true2026.06 | 17.64 | |
| C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine), Entropy-Aware Routing (EAR)=true2026.06 | 17.86 | |
| C-DR MoERouting Strategy=Soft Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine)2026.06 | 18.27 | |
| C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine)2026.06 | 18.43 | |
| C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Single-Stage)2026.06 | 18.52 | |
| C-DR MoERouting Strategy=Hard Routing, Classifier Type=Ground-Truth Routing2026.06 | 18.65 | |
| C-DR MoERouting Strategy=Hard Routing, Classifier Type=Top-layer Classifier (Single-Stage)2026.06 | 18.68 | |
| Single Dataset Fine-tuneFine-tuning Protocol=Projector+LoRA2026.06 | 19.15 | |
| Vanilla-Routing MoERouting Strategy=Utterance-level, Fine-tuning Protocol=Pretrain + Joint2026.06 | 19.55 | |
| Vanilla-Routing MoERouting Strategy=Utterance-level, Fine-tuning Protocol=Pretrain + Gate-only2026.06 | 20.02 | |
| Vanilla-Routing MoERouting Strategy=Utterance-level, Fine-tuning Protocol=Joint2026.06 | 20.3 | |
| Single-ExpertArchitecture=One projector, one LoRA2026.06 | 20.35 | |
| Canary-QwenFine-tuning Protocol=Zero-shot2026.06 | 24.97 |