Relative mismatch improves detection of unusual machine sounds
Relative Mismatch: Local-Reference Calibration of Feature-Space Flows for Anomalous Sound Detection
Sound
Summary
Detecting unusual or abnormal sounds from machines can help spot problems before they get worse. This paper explores a new way to do that by using a method called Relative Mismatch, which compares expected sound features to actual ones in a smart way that focuses on differences from normal sounds nearby. The approach uses a type of model that learns how normal sounds change over time to spot these differences more accurately. Tests on multiple sound datasets show this method works better and more reliably than previous techniques.
What this means in practice
- •For industrial maintenance teams: Monitor equipment sounds to detect early anomalies and prevent failure using a calibrated generative model for more accurate sound analysis.
- •For machine health monitoring vendors: Incorporate a new backend algorithm that outperforms common methods to improve the reliability of sound-based anomaly detection products.$Commercial implications: Enables vendors to sell more accurate and stable anomaly detection solutions for predictive maintenance based on sound.
Authors
Anbai Jiang, Xinhu Zheng, Lvxin Xu, Shuwei Zhang, Wenrui Liang, Pingyi Fan, Wei-Qiang Zhang, Cheng Lu, Jia Liu
Abstract
Anomalous sound detection (ASD) has long been dominated by k-nearest-neighbor (KNN) based detectors, which essentially perform implicit likelihood estimation over normal samples. In this work, we investigate whether generative models can better serve this role. We propose Relative Mismatch, a generative ASD backend powered by flow matching, which learns a velocity field that transports Gaussian noise to a representative feature space of normality. During inference, it measures the mismatch between the oracle and predicted path velocities and aggregates them through a two-level design. To mitigate the inherent mismatch offsets incurred by domain shift, each query is further calibrated with the mismatch of its local normal reference, thereby exposing only its deviation beyond normality. Extensive experiments on DCASE 2020--2025 demonstrate that Relative Mismatch outperforms state-of-the-art backends with the highest score of 71.01, along with strong robustness and training stability. Furthermore, we show that curating a compact and discriminative feature space is the key to unleash the power of generative models for ASD.