Analysis of Feature Representations for Anomalous Sound Detection

TL;DR

This paper evaluates pretrained neural network features for anomalous sound detection, showing that diverse domain-trained features outperform autoencoders, with music-based features often performing best, challenging domain-matching assumptions.

Contribution

It demonstrates the effectiveness of using pretrained neural network features from various domains for anomaly detection, highlighting the superiority over autoencoder baselines.

Findings

Music-based features perform best in most cases.

Pretrained features from different domains outperform autoencoders.

Domain matching between feature extractor and task is less critical than previously thought.

Abstract

In this work, we thoroughly evaluate the efficacy of pretrained neural networks as feature extractors for anomalous sound detection. In doing so, we leverage the knowledge that is contained in these neural networks to extract semantically rich features (representations) that serve as input to a Gaussian Mixture Model which is used as a density estimator to model normality. We compare feature extractors that were trained on data from various domains, namely: images, environmental sounds and music. Our approach is evaluated on recordings from factory machinery such as valves, pumps, sliders and fans. All of the evaluated representations outperform the autoencoder baseline with music based representations yielding the best performance in most cases. These results challenge the common assumption that closely matching the domain of the feature extractor and the downstream task results in better downstream task performance.