List your company
ResearchFormerly WizWand

Automatic Speech Recognition on MyST (test)

8.21WERGumbel-BEARD
7.90249.978712.05514.1313Jun 9, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Gumbel-BEARDBackbone=Whisper-medium, Labeled Data Size=Full2026.068.21
Single Dataset Fine-tuneFine-tuning Protocol=Encoder+Projector+LoRA2026.068.34
Previous SOTA2026.068.5
Gumbel-BEARD (Hard selection)Backbone=Whisper-small, Labeled Data Size=Full2026.068.51
Single Dataset Fine-tuneFine-tuning Protocol=Projector+LoRA2026.068.58
C-DR MoERouting Strategy=Hard Routing, Classifier Type=Ground-Truth Routing2026.068.58
C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Single-Stage)2026.068.58
C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine)2026.068.58
C-DR MoERouting Strategy=Hard Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine), Entropy-Aware Routing (EAR)=true2026.068.58
C-DR MoERouting Strategy=Soft Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine)2026.068.58
C-DR MoERouting Strategy=Soft Routing, Classifier Type=Weighted-layer Classifier (Coarse-to-Fine), Entropy-Aware Routing (EAR)=true2026.068.58
GC-LoRAWhisper model size=medium2026.068.6
GC-LoRAParams=447k, Backbone=Whisper-Medium2026.068.6
C-DR MoERouting Strategy=Hard Routing, Classifier Type=Top-layer Classifier (Single-Stage)2026.068.61
Vanilla-Routing MoEFine-tuning Protocol=Pretrain + Joint2026.068.63
GC-LoRAWhisper model size=large-v32026.068.7
Baseline (BEARD)Backbone=Whisper-small, Labeled Data Size=Full2026.068.73
Vanilla-Routing MoEFine-tuning Protocol=Joint2026.068.74
Gumbel-BEARD (Soft selection)Backbone=Whisper-small, Labeled Data Size=Full2026.068.76
Baseline (SFT)Backbone=Whisper-medium, Labeled Data Size=Full2026.068.86
Gumbel-BEARDBackbone=Whisper-medium, Labeled Data Size=10 h2026.068.88
LoRAWhisper model size=medium2026.068.9
Full FTParams=764M, Backbone=Whisper-Medium2026.068.9
LoRAParams=829k, Backbone=Whisper-Medium2026.068.9
Canary-QwenFine-tuning Protocol=Zero-shot2026.068.96
Gumbel-BEARDBackbone=Whisper-medium, Labeled Data Size=1 h2026.069.15
Baseline (SFT)Backbone=Whisper-medium, Labeled Data Size=10 h2026.069.19
Single-Expert2026.069.27
Baseline (SFT)Backbone=Whisper-small, Labeled Data Size=Full2026.069.34
Gumbel-BEARD (Hard selection)Backbone=Whisper-small, Labeled Data Size=10 h2026.069.35
Baseline (BEARD)Backbone=Whisper-small, Labeled Data Size=10 h2026.069.44
LoRAWhisper model size=large-v32026.069.5
Baseline (SFT)Backbone=Whisper-medium, Labeled Data Size=1 h2026.069.56
GC-LoRAWhisper model size=small2026.069.6
Gumbel-BEARD (Soft selection)Backbone=Whisper-small, Labeled Data Size=10 h2026.069.62
Baseline (PL)Backbone=Whisper-small, Labeled Data Size=Full2026.069.63
LoRAWhisper model size=small2026.069.8
Baseline (SFT)Backbone=Whisper-small, Labeled Data Size=10 h2026.069.94
Gumbel-BEARD (Hard selection)Backbone=Whisper-small, Labeled Data Size=1 h2026.0610.18
Baseline (BEARD)Backbone=Whisper-small, Labeled Data Size=1 h2026.0610.31
Baseline (SFT)Backbone=Whisper-small, Labeled Data Size=1 h2026.0610.64
Gumbel-BEARD (Soft selection)Backbone=Whisper-small, Labeled Data Size=1 h2026.0610.65
Baseline (PL)Backbone=Whisper-small, Labeled Data Size=10 h2026.0611.66
GC-LoRAWhisper model size=base2026.0612.2
Vanilla-Routing MoEFine-tuning Protocol=Pretrain + Gate-only2026.0612.4
Baseline (PL)Backbone=Whisper-small, Labeled Data Size=1 h2026.0612.45
LoRAWhisper model size=base2026.0612.7
Zero-shotParams=0, Backbone=Whisper-Medium2026.0613.1
Zero-shotBackbone=Whisper-medium, Labeled Data Size=no fine-tuning2026.0613.1
Zero-shotBackbone=Whisper-small, Fine-tuning=None2026.0613.4
GC-LoRAWhisper model size=tiny2026.0615.7
LoRAWhisper model size=tiny2026.0615.9