Deep Generic Representations for Domain-Generalized Anomalous Sound Detection

TL;DR

This paper introduces nRep, a domain-generalized anomalous sound detection method using large-scale pre-trained features and kNN, achieving superior performance without fine-tuning or extensive labeled data.

Contribution

It proposes nRep, combining pre-trained feature extractors, MemMixup augmentation, and domain normalization for robust, label-free anomaly detection across domains.

Findings

Outperforms OE-based methods on DCASE2023T2 with 73.79% score

Robust under limited data scenarios

No fine-tuning required for effective domain generalization

Abstract

Developing a reliable anomalous sound detection (ASD) system requires robustness to noise, adaptation to domain shifts, and effective performance with limited training data. Current leading methods rely on extensive labeled data for each target machine type to train feature extractors using Outlier-Exposure (OE) techniques, yet their performance on the target domain remains sub-optimal. In this paper, we present \textit{GenRep}, which utilizes generic feature representations from a robust, large-scale pre-trained feature extractor combined with kNN for domain-generalized ASD, without the need for fine-tuning. \textit{GenRep} incorporates MemMixup, a simple approach for augmenting the target memory bank using nearest source samples, paired with a domain normalization technique to address the imbalance between source and target domains. \textit{GenRep} outperforms the best OE-based approach without a need for labeled data with an Official Score of 73.79\% on the DCASE2023T2 Eval set and demonstrates robustness under limited data scenarios. The code is available open-source.