Automatic EPS Calculation Guided Genetic Algorithm and Incremental PCA Based DBSCAN of Extracted Acoustic Features for Anomalous Sound Detection

This paper focuses on automatic detection of anomalies in audio files. We introduce an automated unsupervised algorithm which integrates AEC-guided Genetic Algorithm and Incremental PCA and DBSCAN to detect the anomaly sound after extracting acoustic features of the normal environments. Technically, Automatic EPS Calculation algorithm (AEC) based Genetic Algorithm optimizes the automatic clustering algorithm’s configuration for Incremental Principal Components Analysis (IPCA) and Density Based Spatial Clustering Algorithm with Noise (DBSCAN) to reduce the number of effective components, which is calculated by guided GA, to a manageable count. The DBSCAN uses the output of the PCA and is able to generate predictions with relatively high accuracy. Besides, because of the challenges when selecting the optimized configuration of the IPCA and DBSCAN especially in complex sounds’ scenarios, this paper also introduces a self-adaptative architecture trying to select an optimized set of parameters based on different test environments. Experiments show that our results are better than the existing methods with less computing costs and more stability. The algorithm is generic and can be applied to detect anomalies in machines so as to give an early warning to maintenance people to avoid serious accidents or disasters.