Hallucination Evaluation on AHA 1.0 (test)
53.8Event OmissionQwen-Audio-AHA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen-Audio-AHAAlignment=DPO (Aligned)2025.12 | 53.8 | 64.1 | 15.9 | 52.6 | |
| Audio Flamingo 32025.12 | 61.4 | 81.6 | 40 | 75.3 | |
| Qwen2.5-OmniAlignment=Base2025.12 | 70.6 | 70.6 | 30.5 | 69.6 | |
| Gemini 2.52025.12 | 74.7 | 80.8 | 34.8 | 58.3 | |
| GPT-4o2025.12 | 77.8 | 72.2 | 34.1 | 63.6 | |
| Kimi-Audio2025.12 | 84.7 | 57.6 | 35.6 | 72.5 |