List your company
ResearchFormerly WizWand

Automatic Speech Recognition on AISHELL-2 (CER)

2.27CERSeed-ASR
1.53886.474411.4116.3456Feb 7, 2026Feb 12, 2026Feb 17, 2026Feb 23, 2026Feb 28, 2026Mar 5, 2026Mar 11, 2026
Updated 7mo ago

Evaluation results

MethodLinks
Seed-ASR2026.032.27
FireRedASR-AED2026.032.52
Kimi-AudioBase Model=Kimi-Audio2026.022.57
Uni-ASRStreaming mode=false, Beam search decoding width=32026.032.6
Qwen3-ASR-1.7B2026.032.71
Fun-ASR-nano2026.032.75
Qwen2-AudioBase Model=Qwen2-Audio2026.023.08
Uni-ASRStreaming mode=true, Chunk length=1000 ms, Beam search decoding width=32026.033.25
Uni-ASRChunk length=10002026.033.25
GLM-ASR-NanoBase Model=GLM-ASR-Nano2026.023.48
SampBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.023.99
SampBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.024.42
SampBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.024.52
CDPrunerBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.024.62
FastAdaSPBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.024.69
MoCha-ASR2026.035.5
A-ToMeBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.025.56
FastAdaSPBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.025.67
CDPrunerBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.025.94
A-ToMeBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.025.95
A-ToMeBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.026.88
FastAdaSPBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.027.02
VisPrunerBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.027.65
CDPrunerBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.029.21
VisPrunerBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.029.67
VisionZipBase Model=Kimi-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.0210.01
VisionZipBase Model=Qwen2-Audio, Token Retention Rate=60%, Compression Ratio=40%2026.0213.85
VisPrunerBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.0214.57
VisionZipBase Model=GLM-ASR-Nano, Token Retention Rate=70%, Compression Ratio=30%2026.0220.55