Unsupervised Anomalous Sound Detection For Machine Condition Using Memory-augmented Deep Autoencoder

Anomalous sound detection (ASD) for machine condition is the task to identify whether the sound emitted from a target machine is normal or anomalous. The idea is to detect unknown anomalous sounds under the condition that only normal sound samples have been provided as training data. The original anomalous sound detection for machine condition uses a deep autoencoder(AE) to learn a normal model that makes anomalous data emerge to more reconstruction error. However, it has been found that sometimes the autoencoder generalizes so well that it can also reconstruct anomalies well. In this paper, we present a memory-augmented deep autoencoder(MemAE) method for machine abnormal sound detection to alleviate the problem. Compared with the traditional AE based method, the accuracy AUC of MemAE in the three machines of Toy-conveyor, Slide rail and Valve are increased by 11%, 3% and 7%.