Anomalous Sound Detection

This repository contains our work for the final project of the course "AI Convergence Capstone Design," offered by Professor Jung-Woo Choi of the Department of Electrical Engineering at KAIST. We tackle DCASE2020 Challenge Task 2: Unsupervised Detection of Anomalous Sounds for Machine Condition Monitoring. This task focuses on developing systems capable of identifying anomalous machine operating sounds without prior exposure to anomalous samples during training.

Datasets

• development dataset
• additional training dataset
• (optional) evaluation dataset

The first two datasets should be merged and organized as follows.

dev_data/
├── fan/
    ├── train/
    ├── test/
├── pump/
    ├── train/
    ├── test/
├── slider/
    ├── train/
    ├── test/
├── ToyCar/
    ├── train/
    ├── test/
├── ToyConveyor/
    ├── train/
    ├── test/
├── valve/
    ├── train/
    ├── test/

Training

To train a model, first specify the path to the dataset (dev_data in the above example) in the root_path of the config.yaml file, and then run:

python train.py

Evaluation

To evaluate the trained model, run:

python eval.py

Results

Machine Type	AUC(%)	pAUC(%)
fan	98.140	94.260
pump	96.541	89.220
slider	99.421	97.048
ToyCar	96.349	89.594
ToyConveyor	82.940	68.800
valve	99.986	99.926
Average	95.563	89.808

References

• S. Choi and J. -W. Choi, "Noisy-Arcmix: Additive Noisy Angular Margin Loss Combined With Mixup For Anomalous Sound Detection," ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Seoul, Korea, Republic of, 2024, pp. 516-520.
• D. Kong, H. Yu and G. Yuan, "Multi-Spectral and Multi-Temporal Features Fusion With SE Network for Anomalous Sound Detection," in IEEE Access, vol. 12, pp. 167262-167277, 2024.