Automated Audio Captioning on AudioCaps (evaluation)
51.8SPIDErSLAM-AAC
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SLAM-AACPre-training Data=AC+CLp+WC+MA2024.10 | 51.8 | 26.8 | 84.1 | 19.4 | 51.5 | 66.8 | |
| SLAM-LLMPre-training Data=AC+CL_P+WC+MA2026.01 | 51.8 | 26.8 | 84.1 | 19.4 | 51.5 | 66.8 | |
| Tang et al.Pre-training Data=AC+CL+WC+LS+GS2024.10 | 50.6 | — | — | — | — | — | |
| Tang et al.Pre-training Data=AC+CL+WC+LS+GS2026.01 | 50.6 | — | — | — | — | — | |
| LOAEPre-training Data=AC+CL+WC2026.01 | 50.5 | 26.7 | 81.6 | 19.3 | 50.4 | 66.2 | |
| EnCLAP-largePre-training Data=AC+CL2024.10 | 49.5 | 25.5 | 80.3 | 18.8 | 49.9 | 65.5 | |
| EnCLAP-largePre-training Data=AC+CL2026.01 | 49.5 | 25.5 | 80.3 | 18.8 | 49.9 | 65.5 | |
| WavCapsPre-training Data=AC+CL+WC2024.10 | 48.5 | 25 | 78.7 | 18.2 | 48.3 | 64.2 | |
| WavCapsPre-training Data=AC+CL+WC2026.01 | 48.5 | 25 | 78.7 | 18.2 | 48.3 | 64.2 | |
| SLAM–AACVariant=LoRA, Enc.=✓, LoRA=✓, Distill=–2026.07 | 41.9 | 23.3 | 66.4 | 17.3 | — | — | |
| SLAM–AACVariant=No LoRA, Enc.=✓, LoRA=✗, Distill=–2026.07 | 38.7 | 23.7 | 59.8 | 17.5 | — | — | |
| CARDVariant=CARD∗, Enc.=✗, LoRA=✓, Distill=Projector (early stages) + LLM2026.07 | 35.2 | 21.2 | 55.4 | 15.1 | — | — | |
| CARDVariant=Proj_Early, Enc.=✗, LoRA=✓, Distill=Early stages→Projector2026.07 | 33.3 | 20.1 | 52.5 | 14.1 | — | — | |
| CARDVariant=CARD⋄, Enc.=✗, LoRA=✓, Distill=Projector (all stages) + LLM2026.07 | 32 | 20.3 | 49.9 | 14.1 | — | — | |
| CARDVariant=No Distill, Enc.=✗, LoRA=✓, Distill=No teacher2026.07 | 27.9 | 18.4 | 43.2 | 12.6 | — | — | |
| CARDVariant=LLM Distill, Enc.=✗, LoRA=✓, Distill=LLM only2026.07 | 27.8 | 18.4 | 43.5 | 12.2 | — | — | |
| CARDVariant=Proj_Full, Enc.=✗, LoRA=✓, Distill=All stages→projector2026.07 | 26.4 | 17.8 | 40.7 | 12.1 | — | — |