List your company

Audio Understanding Robustness on MMAU Sound (SNR=-10dB)

66.83AccuracyEchoDistill
55.629258.537161.44564.3529May 11, 2026

Evaluation results

MethodLinks
EchoDistillBackbone=Qwen2.5-Omni2026.0566.8372.8375.33
STFTBackbone=Step-Audio22026.0563.7570.3369.91
EchoDistillBackbone=Step-Audio22026.0563.6771.7580.08
InitialBackbone=Step-Audio22026.0563.5—70.42
DFLBackbone=Step-Audio22026.0563.571.574.58
WTBackbone=Step-Audio22026.0563.336970.5
EchoDistillBackbone=Qwen2.5-Omni2026.0562.9675.3676.14
EchoDistillBackbone=MiniCPM-o-2.62026.0562.1877.1480.42
InitialBackbone=MiniCPM-o-2.62026.0561.92—77.76
EchoDistillBackbone=MiniCPM-o-2.62026.0561.7979.0781.37
SEENBackbone=MiniCPM-o-2.62026.0561.5677.278.04
DFLBackbone=MiniCPM-o-2.62026.0561.5578.7280.31
SEENBackbone=MiniCPM-o-2.62026.0561.0378.5878.79
InitialBackbone=MiniCPM-o-2.62026.0560.94—79.18
STFTBackbone=MiniCPM-o-2.62026.0560.9377.1177.78
WTBackbone=MiniCPM-o-2.62026.0560.8675.2777.66
STFTBackbone=Qwen2.5-Omni2026.0560.7567.7568.25
DFLBackbone=MiniCPM-o-2.62026.0560.5178.9380.73
STFTBackbone=MiniCPM-o-2.62026.0560.3778.4780.01
SEENBackbone=Step-Audio22026.0559.8367.2577.16
DFLBackbone=Qwen2.5-Omni2026.0559.7564.6667.41
EchoDistillBackbone=Step-Audio22026.0559.5773.6580.32
STFTBackbone=Step-Audio22026.0559.5372.2472.46
STFTBackbone=Qwen2.5-Omni2026.0559.5371.9972.54
InitialBackbone=Step-Audio22026.0559.36—72.75
SEENBackbone=Qwen2.5-Omni2026.0559.3471.8271.6
SEENBackbone=Qwen2.5-Omni2026.0559.3366.2566.42
InitialBackbone=Qwen2.5-Omni2026.0559.32—71.35
WTBackbone=MiniCPM-o-2.62026.0558.5876.1878.01
DFLBackbone=Qwen2.5-Omni2026.0558.2670.0271.35
InitialBackbone=Qwen2.5-Omni2026.0558.17—65.5
DFLBackbone=Step-Audio22026.0558.1470.6571.71
WTBackbone=Qwen2.5-Omni2026.0556.6661.5863.41
WTBackbone=Qwen2.5-Omni2026.0556.466.4868.03
WTBackbone=Step-Audio22026.0556.3467.3869.11
SEENBackbone=Step-Audio22026.0556.0668.2174.91