Anomalous Sound Detection

This chapter provides a comprehensive overview of anomalous sound detection (ASD), a critical technology in audio analysis for security surveillance and industrial applications. Acoustic monitoring offers significant advantages over other methods due to its non-invasive nature, cost-effectiveness, and ability to function without direct line-of-sight. This chapter highlights ASD’s pivotal role in machine condition monitoring, where unsupervised approaches demonstrate superior practicality compared to supervised methods due to the inherent scarcity and high acquisition costs of anomalous samples in real-world scenarios. Various deep learning frameworks for ASD are analyzed, with particular attention to their architectures, learning paradigms, and evaluation metrics. The chapter also addresses significant challenges in this field, including domain generalization, few-shot learning applications, and the integration of pre-trained audio models and audio generation techniques. These research directions aim to enhance ASD’s robustness across varying operational environments and machine types, potentially revolutionizing predictive maintenance, quality control, and safety monitoring across diverse industrial sectors.