List your company

Automatic Speech Recognition on CORAAL (test)

9.25WERGumbel-BEARD
8.44413.884519.32524.7655May 16, 2024Sep 18, 2024Jan 22, 2025May 28, 2025Sep 30, 2025Feb 3, 2026Jun 9, 2026

Evaluation results

MethodLinks
Gumbel-BEARDBackbone=Whisper-medium2026.069.25—
LoRAWhisper model size=large-v32026.069.7—
GC-LoRAWhisper model size=large-v32026.069.7—
Full FTParams=764M, Backbone=Whisper-Medium2026.069.8—
Baseline (SFT)Backbone=Whisper-medium2026.069.81—
GC-LoRAWhisper model size=medium2026.069.9—
GC-LoRAParams=447k, Backbone=Whisper-Medium2026.069.9—
LoRAWhisper model size=medium2026.0610.1—
LoRAParams=829k, Backbone=Whisper-Medium2026.0610.1—
Gumbel-BEARDBackbone=Whisper-small2026.0611.01—
Baseline (SFT)Backbone=Whisper-small2026.0611.7—
GC-LoRAWhisper model size=small2026.0612.3—
LoRAWhisper model size=small2026.0612.7—
Zero-shotParams=0, Backbone=Whisper-Medium2026.0617—
Zero-shotBackbone=Whisper-medium2026.0617.09—
Zero-shotBackbone=Whisper-small2026.0618.69—
GC-LoRAWhisper model size=base2026.0620.1—
LoRAWhisper model size=base2026.0620.3—
ClozeGERBackbone=SpeechGPT, Source Speech=true, LoRA=true, Logits Calibration=true, Post-processing=true2024.0526.78.6
ClozeGERBackbone=LLaMA-2-7b, Source Speech=false, LoRA=true, Logits Calibration=true, Post-processing=true2024.0526.88.2
GC-LoRAWhisper model size=tiny2026.0629—
Whisper Baseline2024.0529.2—
LoRAWhisper model size=tiny2026.0629.4—