Analysis of Feature Representations for Anomalous Sound Detection
This work addresses the problem of improving anomalous sound detection for industrial machinery by challenging the assumption that domain-matched feature extractors are always superior.
This paper evaluates pretrained neural networks as feature extractors for anomalous sound detection, using them to generate semantically rich features for a Gaussian Mixture Model. The study found that music-based representations consistently outperformed an autoencoder baseline and other domain-specific feature extractors.
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.