ToyADMOS2: Another dataset of miniature-machine operating sounds for anomalous sound detection under domain shift conditions

This repository provides a data mixer tool for ToyADMOS2 🚗 🚃, a large-scale dataset for anomaly detection in machine operating sounds (ADMOS) that consist of a large number of operating sounds of miniature machines (toys) under normal and anomaly conditions by deliberately damaging them. You can find the detail of the dataset on the ToyADMOS2 dataset website.

If you find the ToyADMOS2 useful in your work, please consider citing our paper.

@inproceedings{harada2021toyadmos2,
    author = "Harada, Noboru and Niizumi, Daisuke and Takeuchi, Daiki and Ohishi, Yasunori and Yasuda, Masahiro and Saito, Shoichiro",
    title = "{ToyADMOS2}: Another Dataset of Miniature-Machine Operating Sounds for Anomalous Sound Detection under Domain Shift Conditions",
    booktitle = "Proceedings of the 6th Detection and Classification of Acoustic Scenes and Events 2021 Workshop (DCASE2021)",
    address = "Barcelona, Spain",
    month = "November",
    year = "2021",
    pages = "1--5",
    isbn = "978-84-09-36072-7",
    doi. = "10.5281/zenodo.5770113",
    _pdf = {https://dcase.community/documents/workshop2021/proceedings/DCASE2021Workshop_Harada_6.pdf}
}

What is the ToyADMOS2 dataset?

ToyADMOS2 is a unique dataset, which we don't use as it is; it's a set of source material recording samples. We then use a tool provided in this repository, and generate new datasets according to our new recipes.

The samples consist of the recordings under various conditions/configurations for normal/anomaly sounds. You can then edit/program your own set in the recipe so that you can compile new datasets for your research purposes.

A document for how we made it ToyADMOS2_details.pdf is also available. You can also check the detail of anomaly conditions with photos.

Please try making your own!

Download dataset

Visit the ToyADMOS2 dataset website hosted by http://zenodo.org/, and download.

Wanna See the Miniature Machines?

Here're the videos of the toy car and the toy train:

• 🚗 Click to play: ToyCarRun.mp4
• 🚃 Click to play: ToyTrainRun.mp4

Getting Started

Install dependent packages according to the requirements.txt.

This will install essential modules for running tools in this repository.

Example: Making Example Dataset

Run the following will create the equivalent benchmark dataset evaluated in the Table 3 of the paper, which is a compatible file-folder structure with the DCASE2021 challenge task 2. This will create dataset folder your_new_dataset. This will take about an hour.

# This creates `clean` dataset.
python mixer.py /path/to/ToyADMOS2 your_new_dataset recipe_benchmark.xlsx clean
# This creates SNR=6dB dataset.
python mixer.py /path/to/ToyADMOS2 your_new_dataset recipe_benchmark.xlsx 6

• recipe_example_car_shift.xlsx is also another example.
• recipe_template is a template, as well as one more example.

Example: Running Baseline

1. Clone and apply a patch for making evaluation baseline based on dcase2020_task2_baseline.

   git clone https://github.com/y-kawagu/dcase2020_task2_baseline
   cd dcase2020_task2_baseline && patch --binary < ../dcase2020_task2_baseline.patch

2. Make a symbolic link for the baseline that finds data source at dcase2020_task2_baseline/dev_data.

   cd dcase2020_task2_baseline && ln -s ../your_new_dataset dev_data

3. Run the baseline, then you can find the evaluation results stored in result/result.csv.

   cd dcase2020_task2_baseline
   python 00_train.py -d
   python 01_test.py -d

If you find anything missing when running dcase2020_task2_baseline, please follow the instruction in it to install basic modules.

Making Your Dataset

(Example of a recipe file, yes it's an Excel spreadsheet.)

You simply make a copy of template (recipe_template.xlsx), edit yours, then run a tool.

You can find more information in the UsersManual.md.

License

Please check the LICENSE for the detail.

Acknowledgements

The evaluation of this dataset use y-kawagu/dcase2020_task2_baseline. We thank @y-kawagu for your dedication to the DCASE challenges!

This repository is an 2021 version, kudos to @YumaKoizumi for the 2020 efforts of the ToyADMOS-dataset.

References

• Noboru Harada, Daisuke Niizumi, Daiki Takeuchi, Yasunori Ohishi, Masahiro Yasuda, and Shoichiro Saito, "ToyADMOS2: Another dataset of miniature-machine operating sounds for anomalous sound detection under domain shift conditions," 2021
• Yuma Koizumi, Shoichiro Saito, Noboru Harada, Hisashi Uematsu and Keisuke Imoto, "ToyADMOS: A Dataset of Miniature-Machine Operating Sounds for Anomalous Sound Detection," WASPAA, 2019
• Ryo Tanabe, Harsh Purohit, Kota Dohi, Takashi Endo, Yuki Nikaido, Toshiki Nakamura, and Yohei Kawaguchi, "MIMII DUE: Sound Dataset for Malfunctioning Industrial Machine Investigation and Inspection with Domain Shifts due to Changes in Operational and Environmental Conditions," 2021
• DCASE 2020 Challenge Task 2 "Unsupervised Detection of Anomalous Sounds for Machine Condition Monitoring"
  • Baseline system for DCASE 2020 Challenge Task 2 - dcase2020_task2_baseline
• DCASE 2021 Challenge Task 2 "Unsupervised Anomalous Sound Detection for Machine Condition Monitoring under Domain Shifted Conditions"
  • Autoencoder-based baseline system for DCASE2021 Challenge Task 2 - dcase2021_task2_baseline_ae
  • MobileNetV2-based baseline system for DCASE2021 Challenge Task 2 - dcase2021_task2_mobile_net_v2