Transformer-based autoencoder with ID constraint for unsupervised anomalous sound detection

• Published in: EURASIP journal on audio, speech, and music processing Vol. 2023; no. 1; pp. 42 - 16
• Main Authors: Guan, Jian, Liu, Youde, Kong, Qiuqiang, Xiao, Feiyang, Zhu, Qiaoxi, Tian, Jiantong, Wang, Wenwu
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 13.10.2023; Springer Nature B.V; SpringerOpen
• Subjects: Acoustics; AI for Computational Audition: Sound and Music Processing; Anomalies; Anomalous sound detection; Autoencoder; Computation; Constraints; Engineering; Engineering Acoustics; ID classifier; Machine learning; Mathematics in Music; Methodology; Self-supervised learning; Signal,Image and Speech Processing; Sound; Transformers; Weighted anomaly score computation; Anomalous sound detection; ID classifier; Weighted anomaly score computation; Autoencoder
• ISSN: 1687-4722; 1687-4714; 1687-4722
• DOI: 10.1186/s13636-023-00308-4

Unsupervised anomalous sound detection (ASD) aims to detect unknown anomalous sounds of devices when only normal sound data is available. The autoencoder (AE) and self-supervised learning based methods are two mainstream methods. However, the AE-based methods could be limited as the feature learned from normal sounds can also fit with anomalous sounds, reducing the ability of the model in detecting anomalies from sound. The self-supervised methods are not always stable and perform differently, even for machines of the same type. In addition, the anomalous sound may be short-lived, making it even harder to distinguish from normal sound. This paper proposes an ID-constrained Transformer-based autoencoder (IDC-TransAE) architecture with weighted anomaly score computation for unsupervised ASD. Machine ID is employed to constrain the latent space of the Transformer-based autoencoder (TransAE) by introducing a simple ID classifier to learn the difference in the distribution for the same machine type and enhance the ability of the model in distinguishing anomalous sound. Moreover, weighted anomaly score computation is introduced to highlight the anomaly scores of anomalous events that only appear for a short time. Experiments performed on DCASE 2020 Challenge Task2 development dataset demonstrate the effectiveness and superiority of our proposed method.