File size: 1,485 Bytes
f964c91
 
 
 
4bd26a9
 
cde039c
 
 
 
 
 
 
 
59f4685
cde039c
 
 
 
c4100fa
 
cde039c
 
 
 
 
4bd26a9
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
---
pretty_name: DataSynthSELD
size_categories:
- 100B<n<1T
task_categories:
- audio-classification
---

# PSELDNets: Pre-trained Neural Networks on Large-scale Synthetic Datasets for Sound Event Localization and Detection

- [Paper](https://arxiv.org/abs/2411.06399)
- [GitHub](https://github.com/Jinbo-Hu/PSELDNets)

1. This repo contains 67,000 1-minute clips amounting to approximately 1,117 hours for training, and 3,060 1-minute clips amounting to roughly 51 hours for testing. 
2. These datasets features an ontology of 170 sound classes and are generated by convolving sound event clips from [FSD50K](https://zenodo.org/records/4060432) with simulated SRIRs (for training) or collected SRIRs from [TAU-SRIR DB](https://zenodo.org/records/6408611) (for testing).
3. The datasets are generated by this [tools](https://github.com/Jinbo-Hu/SELD-Data-Generator).


## Download
- [Synthetic Datasets](https://huggingface.co/datasets/Jinbo-HU/PSELDNets/tree/main/dataset)
- [Pre-trained checkpoints](https://huggingface.co/datasets/Jinbo-HU/PSELDNets/tree/main/model)

## Citation
Please cite our papers as below if you use the datasets, codes and models of PSELDNets.


[1] Jinbo Hu, Yin Cao, Ming Wu, Fang Kang, Feiran Yang, Wenwu Wang, Mark D. Plumbley, Jun Yang, "PSELDNets: Pre-trained Neural Networks on Large-scale Synthetic Datasets for Sound Event Localization and Detection" [arXiv:2411.06399](https://arxiv.org/abs/2411.06399), 2024. [URL](https://arxiv.org/abs/2411.06399)