Retaining Mixture Representations for Domain Generalized Anomalous Sound Detection

2025-10-29

AI Summary

Distribution shifts—such as unknown low signal-to-noise ratios and heterogeneous noise types—in real-world anomalous sound detection (ASD) degrade model generalization and distort learned representations. Method: We propose a “preserve-rather-than-denoise” training paradigm built upon a frozen self-supervised audio encoder. Our approach introduces a hybrid embedding alignment mechanism: using teacher representations derived from convex combinations of clean sources and noise as supervision, it jointly optimizes multi-label classification loss and hybrid alignment loss to guide the student model toward robust, consistent representations of mixed acoustic sources. Inference requires no additional adaptation, preserving efficiency. Results: Experiments demonstrate substantial improvements in out-of-distribution generalization under static/non-static noise and noise-mismatch conditions. The method effectively narrows the gap between learned and ideal mixed-source representations, offering a scalable, highly robust solution for ASD in realistic acoustic environments.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Search and Optimization: Mixed Discrete/Continuous Search
• Natural Language Processing: Learning & Optimization for NLP

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Anomalous sound detection (ASD) in the wild requires robustness to distribution shifts such as unseen low-SNR input mixtures of machine and noise types. State-of-the-art systems extract embeddings from an adapted audio encoder and detect anomalies via nearest-neighbor search, but fine tuning on noisy machine sounds often acts like a denoising objective, suppressing noise and reducing generalization under mismatched mixtures or inconsistent labeling. Training-free systems with frozen self-supervised learning (SSL) encoders avoid this issue and show strong first-shot generalization, yet their performance drops when mixture embeddings deviate from clean-source embeddings. We propose to improve SSL backbones with a retain-not-denoise strategy that better preserves information from mixed sound sources. The approach combines a multi-label audio tagging loss with a mixture alignment loss that aligns student mixture embeddings to convex teacher embeddings of clean and noise inputs. Controlled experiments on stationary, non-stationary, and mismatched noise subsets demonstrate improved robustness under distribution shifts, narrowing the gap toward oracle mixture representations.

Problem

• Detecting anomalous sounds under domain distribution shifts
• Preserving mixture representations instead of denoising sounds
• Improving generalization with mixture alignment and multi-label training

Innovation

• Retain-not-denoise strategy preserves mixed sound information
• Multi-label audio tagging loss with mixture alignment
• Aligns student mixture embeddings to convex teacher embeddings