SSAD: source separation followed by anomaly detection

This is an official repository for our paper, "Activity-informed Industrial Audio Anomaly Detection via Source Separation".

If you are considering using repository, please cite our paper:

@inproceedings{kim2023activity,
  title={Activity-informed Industrial Audio Anomaly Detection via Source Separation},
  author={Jaechang Kim and Yunjoo Lee and Hyun Mi Cho and Dong Woo Kim and Chi Hoon Song and Jungseul Ok},
  booktitle={IEEE International Conference on Acoustics, Speech and Signal Processing},
  year={2023}
}

Environment Setting

conda env create -n asteroid
conda activate asteroid
conda install python=3.7
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch-lts -c nvidia

# for asteroid
pip install -r requirements/dev.txt
pip install -e .
pip install torchmetrics==0.6.0

# for anomaly detection
pip install -r anomaly/requirements.txt

Training Source Separation Models

To run X-UMX change the configuration file considering the type of data and the use of control signal.

1. First change the configuration

cd informed-X-UMX
vi local/conf_???.yml

Edit local/conf_base.yml for XUMX baseline and local/conf_informed.yml for informed source separation model.

• data:train_dir -> MIMII dataset directory
• data:output -> directory where checkpoint and log files will be saved
• data:machine_type -> machine types to use
• data:sources -> machine ids to use
• model:pretrained -> pretrained model path

2. train the model by running

cd informed-X-UMX
train.py --conf local/conf_base.yml

Run train.py for with given configuration file.

Anomaly Detection models

Edit anomaly/baseline.yaml

• base_directory -> MIMII dataset path

Oracle baseline

Edit anomaly/baseline.py

• Check datapath near line 196
  • dirs = sorted(glob.glob(os.path.abspath("{base}/6dB/valve/id_00".format(base=param["base_directory"]))))
  • Choose which machines (types, id) to use

cd anomaly
python baseline.py

Mixture baseline

Edit anomaly/baseline_mix.py

• Check datapath near line 228
  • dirs = sorted(glob.glob(os.path.abspath("{base}/6dB/valve/id_00".format(base=param["base_directory"]))))
  • Choose which machines (types, id) to use
• Check machine_types near line 42
  • Those machine types will be used to make a mixture

cd anomaly
python baseline_mix.py

SSAD (Proposed Method)

Edit anomaly/baseline_src_xumx_original.py

• Check datapath near line 318
  • dirs = sorted(glob.glob(os.path.abspath("{base}/6dB/valve/id_00".format(base=param["base_directory"]))))
  • Choose which machines (types, id) to use
• Check trained separation model path near 363
• Check conf near line 43
  • S1, S2 -> machine id
  • FILE -> AE model path (to save)

cd anomaly
python baseline_src_xumx_original.py

Acknowledgement

This repository is based on

• https://github.com/asteroid-team/asteroid
• https://github.com/MIMII-hitachi/mimii_baseline