Temporal Sound Event Localization on Controlled Sound Event Sequence Benchmark
98.91Overlap (Begin)Audio Flamingo 3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Audio Flamingo 3Model=Audio Flamingo 3 (AF3), Prompting=Direct (explicit timestamp output)2026.06 | 98.91 | 33.46 | 7.87 | 46.75 | |
| Audio Flamingo 3 (Window)Model=Audio Flamingo 3 (AF3), Technique=Instruction-based vector steering2026.06 | 73.98 | 48.31 | 83.87 | 68.72 | |
| Qwen2-Audio (Window)Model=Qwen2-Audio, Technique=Instruction-based vector steering2026.06 | 36.9 | 73.39 | 72.32 | 60.87 | |
| Qwen2-AudioModel=Qwen2-Audio, Prompting=Direct (explicit timestamp output)2026.06 | 36.86 | 53.04 | 5.61 | 31.84 | |
| RandomDescription=Places the window uniformly at random across valid positions2026.06 | 20.81 | 41.61 | 20.81 | 27.74 |