Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Harmfulness Evaluation on SORRY-Bench audio
Loading...
78.41
ASR Accuracy
MDSteer-c2r
-2.8972
18.2114
39.32
60.4286
Oct 20, 2025
ASR Accuracy
Updated 26d ago
Evaluation Results
Method
Method
Links
ASR Accuracy
MDSteer-c2r
Model=Qwen2-Audio
2025.10
78.41
MDSteer-h2s
Model=Qwen2-Audio
2025.10
75.45
MDSteer-h2s
Model=Kimi-Audio
2025.10
55
RRS
Defense runtime (s)=1145
2025.10
38.41
No Defense
Model=Qwen2-Audio
2025.10
27.5
No Defense
2025.10
27.5
MDSteer-c2r
Model=Kimi-Audio
2025.10
21.59
AdaShield
Model=Qwen2-Audio
2025.10
20.45
SARSteer
Model=Qwen2-Audio
2025.10
13.41
SARSteer
Defense runtime (s)=266
2025.10
13.41
No Defense
Model=Kimi-Audio
2025.10
12.5
FSD
Model=Kimi-Audio
2025.10
11.14
FSD
Model=Qwen2-Audio
2025.10
10.55
SARSteer
Model=Kimi-Audio
2025.10
6.14
AdaShield
Model=Kimi-Audio
2025.10
0.23
Feedback
Search any
task
Search any
task