This repository is a recipe for running the second-place method in Task 2 of the DCASE 2022 competition for the performance of anomalous sound detection systems.
The method consists of two stages: a feature extractor that utilizes pseudo-anomalous data and an anomalous detector.
Details of the method are written in our Technical Report.
We presented our original proposed method at EUSIPCO 2022.
- Python 3.9+
- Cuda 11.3
$ git clone https://github.com/ibkuroyagi/dcase2022_task2_challenge_recipe.git
$ cd dcase2022_task2_challenge_recipe/tools
$ make- dcase2022-task2: The main challenge of this task is to detect unknown anomalous sounds under the condition that only normal sound samples have been provided as training data.
To run the recipe, please follow the below instruction.
# Let us move on the recipe directory
$ cd scripts
# Run the recipe from scratch
$ ./job.sh
# You can change config via command line
$ ./job.sh --no <the_number_of_your_customized_yaml_config>
# You can select the stage to start and stop
$ ./job.sh --stage 1 --start_stage 3
# After all machine types have completed Stage 5, starting Stage 2.
# You can see the results at exp/all/**/score*.csv
$ ./job.sh --stage 2
# If you would like to ensemble several models, please following commands.
$ ./domain_classifier_job.sh
$ . ./path.sh
$ python ./local/get_domain_classifier_weight.py
$ python ./local/domain_generalization_ave.py
If you think this work is useful to you, please cite:
@inproceedings{kuroyanagi2022eusipco,
title={{Improvement of Serial Approach to Anomalous Sound Detection by Incorporating Two Binary Cross-Entropies for Outlier Exposure}},
author={Ibuki Kuroyanagi and Tomoki Hayashi and Kazuya Takeda and Tomoki Toda},
booktitle={2022 30th European Signal Processing Conference (EUSIPCO)},
pages={294--298},
year={2022},
organization={IEEE}
}
@techreport{Kuroyanagi2022dcase,
Author = "Kuroyanagi, Ibuki and Hayashi, Tomoki and Takeda, Kazuya and Toda, Tomoki",
title = "Two-stage anomalous sound detection systems using domain generalization and specialization techniques",
institution = "DCASE2022 Challenge",
year = "2022",
}
Ibuki Kuroyanagi (@ibkuroyagi)
E-mail: kuroyanagi.ibuki<at>g.sp.m.is.nagoya-u.ac.jp