Exploring sound source separation for acoustic condition monitoring in industrial scenarios

摘要

This paper evaluates the application of three methods for Sound Source Separation (SSS) in industrial acoustic condition monitoring scenarios. To evaluate the impact of SSS, we use a machine learning approach where a classifier is trained to detect a specific operating machine. The evaluation procedure is based on simulated and measured data, comprising three different machine sounds as targets and 10 interfering signals. Various intermixing levels of target and interfering signal are taken into account, using three different signal-to-interference ratios. Results show that the chosen source separation methods, originally developed for music analysis, work well for industrial signals, significantly improving the classification accuracy.