Skip to content

MihawkHu/DCASE2020_task1

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

DCASE2020 Task1

Task1a | Task1b | Video | dcase paper | icassp paper | Keras | TensorFlow

New We add a list on recent related ASC works containing discussion with this open resource ASC studies. Welcome to open an issue for adding related reference with the open resouce studies or just sharing your work.

This work has been accepted to IEEE ICASSP 2021! (Session Time: Friday, 11 June, 13:00 - 13:45 presented by Hu Hu)

Introduction

This is an implementation of DCASE 2020 Task 1a and DCASE 2020 Task 1b on Acoustic Scene Classification with Multiple Devices. We attain 2nds for both Task-1a and Task-1b in the official challenge 2020. Technical Report.

We sincerely thank all the team members and advisors from Georgia Tech ECE, Tencent Media Lab, USTC, and Univeristy of Enna.

Experimental results

Task 1a

Tested on DCASE 2020 task 1a development data set. The train-test split way follows the official recomendation.

System Dev Acc.
Official Baseline 51.4%
10-class FCNN 76.9%
10-class Resnet 74.6%
10-class fsFCNN 76.2%
Two-stage ensemble system 81.9%

Task 1b

Tested on DCASE 2020 task 1b development data set. The train-test split way follows the official recomendation.

System Dev Acc. (size)
Original model
Dev Acc. (size)
Quantization
Official Baseline 87.3% (450K) -
Mobnet 95.2% (3.2M) 94.8% (411K)
small-FCNN 96.4% (2.8M) 96.3% (357K)
Mobnet + small-FCNN-v1 96.8% (1.8M+1.9M) 96.7% (497K)
small-FCNN-v1 + small-FCNN-v2 96.5% (1.9M+2.1M) 96.3% (499K)

How to use

Task 1a

Please refer to the README.md in ./task1a/ for detailed instructions.

Task 1b

Please refer to the README.md in ./task1b/ for detailed instructions.

Pre-trained models

  • Pre-trained keras models are provided in ./task1a/3class/pretrained_models/, task1a/10class/pretrained_models/, and ./task1b/pretrained_models/. You can get reported results by directly evaluate pre-trained models.

  • tensorflow >= 2.0 pretrained models. We also provide some pretrained DCASE task1 models in tensorflow >= 2.0 format. Please refer to ./other_TF2_format_pretrained/

Reference

If this work helps or has been used in your research, please consider to cite both papers below. Thank you!

@inproceedings{hu2021two,
  title={A two-stage approach to device-robust acoustic scene classification},
  author={Hu, Hu and Yang, Chao-Han Huck and Xia, Xianjun and Bai, Xue and Tang, Xin and Wang, Yajian and Niu, Shutong and Chai, Li and Li, Juanjuan and Zhu, Hongning and others},
  booktitle={ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={845--849},
  year={2021},
  organization={IEEE}
}


@misc{hu2020devicerobust,
    title={Device-Robust Acoustic Scene Classification Based on Two-Stage Categorization and Data Augmentation},
    author={Hu Hu and Chao-Han Huck Yang and Xianjun Xia and Xue Bai and Xin Tang and Yajian Wang and Shutong Niu and Li Chai and Juanjuan Li and Hongning Zhu and Feng Bao and Yuanjun Zhao and Sabato Marco Siniscalchi and Yannan Wang and Jun Du and Chin-Hui Lee},
    year={2020},
    eprint={2007.08389},
    archivePrefix={arXiv},
    primaryClass={eess.AS}
}

Acknowledgements

Codes borrows heavily from DCASE2019-Task1 and dcase2020_task1_baseline. We appreciate the researchers contributing to this ASC community.