Handling Domain Shifts for Anomalous Sound Detection: A Review

When detecting anomalous sounds in possibly complex environments, one of the main difficulties is that trained models need to be sensitive to subtle differences of monitored target signals. At the same time, for many practical applications these models should be insensitive to changes of the acoustic domain. Examples of these domain shifts are changing the microphone or the location of acoustic sensors, which both may have a much stronger impact on the acoustic signal than subtle anomalies. Moreover, users want to train a model only with a relatively large collection of source domain data. Furthermore, such a trained model should be able to generalize well to any unseen domain by only providing very few samples of the target domain to define how acoustic signals in this domain sound like. In this work, we review and discuss recent publications focusing on this domain generalization problem for anomalous sound detection in the context of the DCASE challenges on acoustic machine condition monitoring.