Audio Scene Generation on ASG-Bench Paired subset
93.62Event (%)Reference Audio
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Reference Audio2026.06 | 93.62 | 95.26 | 4.25 | 4.54 | 4.26 | |
| Audio-OscarLLM=DeepSeek-V4-Flash, Thinking=false2026.06 | 92.14 | 93.36 | 4.13 | 4.23 | 3.96 | |
| Audio-OscarLLM=DeepSeek-V4-Flash, Thinking=true2026.06 | 92.04 | 93.78 | 4.03 | 4.2 | 3.88 | |
| Audio-OscarLLM=Qwen-397B-A17B, Thinking=false2026.06 | 91.31 | 93.72 | 4.11 | 4.2 | 3.95 | |
| Audio-OscarLLM=Qwen-122B-A10B, Thinking=false2026.06 | 90.25 | 93.56 | 4.18 | 4.31 | 4.01 | |
| WavJourneyLLM=DeepSeek-V4-Flash, Thinking=false2026.06 | 87.26 | 92.09 | 3.84 | 3.9 | 3.57 | |
| Any2Speech2026.06 | 80.38 | 84.54 | 4.17 | 3.82 | 3.78 |