Improving Anomalous Sound Detection with Attribute-aware Representation from Domain-adaptive Pre-training

2025-09-16

AI Summary

Anomaly sound detection (ASD) faces a critical bottleneck in practice due to the scarcity and high cost of machine-attribute annotations (e.g., device type, operating condition). Method: This paper proposes a weakly supervised ASD framework that eliminates the need for ground-truth attribute labels. It first leverages a domain-adaptive pre-trained model to extract robust acoustic representations; then applies agglomerative hierarchical clustering to automatically generate machine-attribute pseudo-labels; finally, fine-tunes the model using these pseudo-labels as supervision. Contribution/Results: To the best of our knowledge, this is the first work to integrate domain-adaptive pre-training with unsupervised clustering for pseudo-attribute labeling in ASD, substantially reducing reliance on manual annotation. Evaluated on the DCASE 2025 Challenge dataset, the method outperforms all prior state-of-the-art systems, demonstrating superior generalization and establishing a novel paradigm for weakly supervised ASD.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

Anomalous Sound Detection (ASD) is often formulated as a machine attribute classification task, a strategy necessitated by the common scenario where only normal data is available for training. However, the exhaustive collection of machine attribute labels is laborious and impractical. To address the challenge of missing attribute labels, this paper proposes an agglomerative hierarchical clustering method for the assignment of pseudo-attribute labels using representations derived from a domain-adaptive pre-trained model, which are expected to capture machine attribute characteristics. We then apply model adaptation to this pre-trained model through supervised fine-tuning for machine attribute classification, resulting in a new state-of-the-art performance. Evaluation on the Detection and Classification of Acoustic Scenes and Events (DCASE) 2025 Challenge dataset demonstrates that our proposed approach yields significant performance gains, ultimately outperforming our previous top-ranking system in the challenge.

Problem

• Detects anomalous sounds without machine attribute labels
• Uses domain-adaptive pre-training for pseudo-attribute assignment
• Improves performance on acoustic scene detection tasks

Innovation

• Agglomerative hierarchical clustering for pseudo-labels
• Domain-adaptive pre-trained model adaptation
• Supervised fine-tuning for attribute classification