List your company

Sound Event Detection on DESED (test)

58.4Event-based F1Qwen2-Audio (TimePro-RL)
1.216.0530.945.75Apr 15, 2026

Evaluation results

MethodLinks
Qwen2-Audio (TimePro-RL)Scale=7B, Evaluation Protocol=Post-Trained with TimePro-RL2026.0458.4——————
Qwen2.5-Omni (TimePro-RL)Scale=7B, Evaluation Protocol=Post-Trained with TimePro-RL2026.0457.6——————
Kimi-AudioScale=7B, Evaluation Protocol=Finetuned2026.0450.9——————
Qwen2-AudioScale=7B, Evaluation Protocol=Finetuned2026.0449.8——————
Qwen2.5-OmniScale=7B, Evaluation Protocol=Finetuned2026.0448.9——————
Qwen2.5-OmniScale=7B, Evaluation Protocol=Zero-shot2026.0413.7——————
Audio-Flaming2Scale=3B, Evaluation Protocol=Finetuned2026.048.9——————
Qwen2-AudioScale=7B, Evaluation Protocol=Zero-shot2026.043.4——————
Audio Flamingo 32026.04—————51.853.7
Kimi-Audio2026.04—————44.1
Qwen2-Audio2026.04—————28.233.8
SpotSound-A2026.04—————5857.8
SpotSound-Q2026.04—————66.661.1
System #1Separation=false, Text Query=None2024.11—0.50.7741.274———
System #2 (Ground Truth)Separation=true, Text Query=Ground Truth2024.11—0.440.7231.163———
System #3 (All Event Texts)Separation=true, Text Query=All Event Texts2024.11—0.4310.7011.132———
System #4 (Ours w/o F.T.)Separation=true, Text Query=Ours w/o F.T.2024.11—0.4370.7061.143———
System #5 (Ours w/o LLM)Separation=true, Text Query=Ours w/o LLM2024.11—0.4110.6691.08———
System #6 (Our w/o CL)Separation=true, Text Query=Our w/o CL2024.11—0.4320.7021.134———
System #7 (Ours)Separation=true, Text Query=Ours2024.11—0.4330.7031.136———
TimeAudio2026.04—————419.3