List your company
ResearchFormerly WizWand

Automatic Speech Recognition on VoxPopuli (dev)

6.02WERASR + DLM
5.83487.08498.3359.5851Dec 15, 2025Jan 5, 2026Jan 26, 2026Feb 16, 2026Mar 9, 2026Mar 30, 2026Apr 20, 2026
Updated 6mo ago

Evaluation results

MethodLinks
ASR + DLMSystem=ASR + DLM2025.126.02
NIM4-ASRModel Size=2.3B, Inference Mode=Offline2026.046.18
OursModel Size=2.3B2026.046.25
NIM4-ASRModel Size=2.3B, Inference Mode=Stream2026.046.26
ASR + LMSystem=ASR + LM2025.126.3
ASR onlySystem=ASR only2025.126.61
Qwen3-Omni-InstModel Size=30B-A3B (↑)2026.046.86
Qwen3-Omni InstructModel Size=30B-A3B, Inference Mode=Offline2026.046.86
Qwen3-ASR-1.7BModel Size=2.0B (↓)2026.047.58
Qwen3-ASR 1.7BModel Size=2.0B, Inference Mode=Offline2026.047.58
Fun-ASR-nanoModel Size=0.8B (↓)2026.047.86
Fun-ASR NanoModel Size=0.8B, Inference Mode=Offline2026.047.86
GLM-ASR-nanoModel Size=1.5B (↓)2026.048.78
GLM-ASR NanoModel Size=1.5B, Inference Mode=Offline2026.048.78
Step-Audio2-miniModel Size=8B+ (↑)2026.048.86
Step-Audio2 MiniModel Size=8B+, Inference Mode=Offline2026.048.86
FireRedASR2S LLMModel Size=8B+, Inference Mode=Offline2026.049.38
FireRedASR-LLMModel Size=8B+ (↑)2026.0410.65