Distributed collaborative anomalous sound detection by embedding sharing

TL;DR

The paper addresses privacy-preserving anomalous sound detection (ASD) across multiple clients with non-identical data distributions by sharing embeddings from a common pre-trained audio model and performing outlier exposure on aggregated embeddings. A server-side ID-classification model is trained on these embeddings, after which clients refine an anomaly detector using transformed embeddings, enabling cross-client learning without exposing raw data. Experiments on the DCASE2020 Task2 dataset show that embedding-sharing with OpenL3 embeddings achieves the best overall performance, with an average ASD AUC improvement of $6.8\%$ over locally trained unsupervised baselines, demonstrating robustness to extreme non-IID data. The approach offers a practical privacy-preserving alternative to federated and split learning for collaborative ASD in industrial settings, with implications for scalable and secure machine health monitoring.$6.8\%$

Abstract

To develop a machine sound monitoring system, a method for detecting anomalous sound is proposed. In this paper, we explore a method for multiple clients to collaboratively learn an anomalous sound detection model while keeping their raw data private from each other. In the context of industrial machine anomalous sound detection, each client possesses data from different machines or different operational states, making it challenging to learn through federated learning or split learning. In our proposed method, each client calculates embeddings using a common pre-trained model developed for sound data classification, and these calculated embeddings are aggregated on the server to perform anomalous sound detection through outlier exposure. Experiments showed that our proposed method improves the AUC of anomalous sound detection by an average of 6.8%.