# Multi-label Sound Event Classification on AS 20K

Source: [Multi-label Sound Event Classification on AS 20K](<https://www.sota2.com/research/sota/multi-label-sound-event-classification-on-as-20k>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-25T08:03:22.571Z

Evaluation of multi-label sound event classification performance on the AS 20K dataset, measured using mAP.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Multi-label sound event classification |
| Dataset (including any stated split) | AS 20K |
| Metric | mAP |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 17 of 17 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [mAP](<https://www.sota2.com/research/sota/multi-label-sound-event-classification-on-as-20k.md?metric=mAP>): Higher is better; unit: PERCENT.

## Recorded results

| Method | mAP | Result date | Report paper |
| --- | --- | --- | --- |
| EquiAV | 42.4 | Not recorded | 6a3cb63e206542251d7ca6ca |
| MAViL | 41.8 | Not recorded | 6a3cb63e206542251d7ca6ca |
| MJEPA ViT-g + data scaling | 40.9 | Not recorded | 6a3cb63e206542251d7ca6ca |
| SSLAM | 40.9 | Not recorded | 6a3cb63e206542251d7ca6ca |
| MJEPA ViT-L + data scaling | 40 | Not recorded | 6a3cb63e206542251d7ca6ca |
| SPEAR | 39.4 | Not recorded | 6a3cb63e206542251d7ca6ca |
| MJEPA ViT-L | 38.9 | Not recorded | 6a3cb63e206542251d7ca6ca |
| AJEPA | 38.4 | Not recorded | 6a3cb63e206542251d7ca6ca |
| CAV-MAE | 37.7 | Not recorded | 6a3cb63e206542251d7ca6ca |
| EquiAV | 34.3 | Not recorded | 6a3cb63e206542251d7ca6ca |
| Dasheng-1.2B | 31.6 | Not recorded | 6a3cb63e206542251d7ca6ca |
| SSLAM | 31.4 | Not recorded | 6a3cb63e206542251d7ca6ca |
| MAViL | 30 | Not recorded | 6a3cb63e206542251d7ca6ca |
| CAV-MAE Sync | 21.7 | Not recorded | 6a3cb63e206542251d7ca6ca |
| CAV-MAE | 19.4 | Not recorded | 6a3cb63e206542251d7ca6ca |
| AJEPA | 18 | Not recorded | 6a3cb63e206542251d7ca6ca |
| SPEAR | 11.9 | Not recorded | 6a3cb63e206542251d7ca6ca |

### EquiAV

mAP: 42.4. Result ID: 6a3ce0b8fbc373592c51efe3.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73c31e209131cea5e53c.md>)

[Implementation](<https://github.com/JongSuk1/EquiAV>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"170M","Modality":"A+V","Pre-train data":"AS"}. This record does not assert independent reproduction.

### MAViL

mAP: 41.8. Result ID: 6a3ce0b8fbc373592c51efe2.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa700d1e209131cea48e29.md>)

[Implementation](<https://github.com/facebookresearch/MAViL>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"170M","Modality":"A+V","Pre-train data":"IN+AS"}. This record does not assert independent reproduction.

### MJEPA ViT-g + data scaling

mAP: 40.9. Result ID: 6a3ce0b8fbc373592c51efdd.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"1B","Modality":"A+V","Pre-train data":"AS+VM2M"}. This record does not assert independent reproduction.

### SSLAM

mAP: 40.9. Result ID: 6a3ce0b8fbc373592c51efdf.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79741e209131cea7db8d.md>)

[Implementation](<https://github.com/ta012/SSLAM>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"88M","Modality":"A","Pre-train data":"AS"}. This record does not assert independent reproduction.

### MJEPA ViT-L + data scaling

mAP: 40. Result ID: 6a3ce0b8fbc373592c51efdc.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"300M","Modality":"A+V","Pre-train data":"AS+VM2M"}. This record does not assert independent reproduction.

### SPEAR

mAP: 39.4. Result ID: 6a3ce0b8fbc373592c51efe0.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/690442703ef7ccd74e5e4ac8.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"600M","Modality":"A","Pre-train data":"197k hrs mix"}. This record does not assert independent reproduction.

### MJEPA ViT-L

mAP: 38.9. Result ID: 6a3ce0b8fbc373592c51efdb.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"300M","Modality":"A+V","Pre-train data":"AS"}. This record does not assert independent reproduction.

### AJEPA

mAP: 38.4. Result ID: 6a3ce0b8fbc373592c51efde.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71d41e209131cea533c4.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"85M","Modality":"A","Pre-train data":"AS"}. This record does not assert independent reproduction.

### CAV-MAE

mAP: 37.7. Result ID: 6a3ce0b8fbc373592c51efe1.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f041e209131cea4345b.md>)

[Implementation](<https://github.com/yuangongnd/cav-mae>)

Recorded evaluation configuration: {"Evaluation protocol":"Full-model finetuning","Params":"170M","Modality":"A+V","Pre-train data":"IN+AS"}. This record does not assert independent reproduction.

### EquiAV

mAP: 34.3. Result ID: 6a3ce0b8fbc373592c51efd9.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73c31e209131cea5e53c.md>)

[Implementation](<https://github.com/JongSuk1/EquiAV>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"170M","Modality":"A+V","Pre-train data":"AS"}. This record does not assert independent reproduction.

### Dasheng-1.2B

mAP: 31.6. Result ID: 6a3ce0b8fbc373592c51efd6.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73931e209131cea5d51d.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"1.2B","Modality":"A","Pre-train data":"272k hrs mix"}. This record does not assert independent reproduction.

### SSLAM

mAP: 31.4. Result ID: 6a3ce0b8fbc373592c51efd4.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79741e209131cea7db8d.md>)

[Implementation](<https://github.com/ta012/SSLAM>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"88M","Modality":"A","Pre-train data":"AS"}. This record does not assert independent reproduction.

### MAViL

mAP: 30. Result ID: 6a3ce0b8fbc373592c51efd8.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa700d1e209131cea48e29.md>)

[Implementation](<https://github.com/facebookresearch/MAViL>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"170M","Modality":"A+V","Pre-train data":"IN+AS"}. This record does not assert independent reproduction.

### CAV-MAE Sync

mAP: 21.7. Result ID: 6a3ce0b8fbc373592c51efda.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78b41e209131cea79dbd.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"170M","Modality":"A+V","Pre-train data":"AS"}. This record does not assert independent reproduction.

### CAV-MAE

mAP: 19.4. Result ID: 6a3ce0b8fbc373592c51efd7.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6f041e209131cea4345b.md>)

[Implementation](<https://github.com/yuangongnd/cav-mae>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"170M","Modality":"A+V","Pre-train data":"IN+AS"}. This record does not assert independent reproduction.

### AJEPA

mAP: 18. Result ID: 6a3ce0b8fbc373592c51efd3.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71d41e209131cea533c4.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"85M","Modality":"A","Pre-train data":"AS"}. This record does not assert independent reproduction.

### SPEAR

mAP: 11.9. Result ID: 6a3ce0b8fbc373592c51efd5.

[Reporting paper](<https://www.sota2.com/research/paper/6a3cb63e206542251d7ca6ca.md>)

[Method origin paper](<https://www.sota2.com/research/paper/690442703ef7ccd74e5e4ac8.md>)

Recorded evaluation configuration: {"Evaluation protocol":"Frozen feature probing","Params":"600M","Modality":"A","Pre-train data":"197k hrs mix"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
