SSDPT: Self-supervised dual-path transformer for anomalous sound detection

Abstract

Anomalous sound detection for machine condition monitoring or structural health monitoring is essential in the development of Industry 4.0. However, the anomalous sounds of machines are unpredictable and are hard to collect in real-world factories. Therefore, the anomalous sound detection methods have to learn robust acoustic representations under the situation that only normal sounds are provided, and effectively detect the anomalous sounds while being applied. In this article, we propose a self-supervised dual-path Transformer (SSDPT) network, which is purely based on attention modules, to detect anomalous sounds for predictive maintenance of the machine. The SSDPT network splits the acoustic features into segments and employs several DPT blocks for time and frequency modeling. DPT blocks use self-attention modules to alternately model the interactive information about the frequency and temporal components of the segmented acoustic features. To address the problem of lack of anomalous sound, we adopt a self-supervised learning approach to train the network with normal sound. Specifically, this approach randomly masks and reconstructs the acoustic features, and jointly classifies machine identity information to improve the performance of anomalous sound detection. We evaluated our method on the DCASE2021 task2 dataset. The experimental results show that the SSDPT network increases in the harmonic mean AUC score compared with state-of-the-art methods of anomalous sound detection.