Improve the accuracy of unsupervised abnormal sound detection using a small number of abnormal sound data. The threshold value determination unit (13) calculates the degree of abnormality for each of a plurality of abnormal sound data using the normal model learned using the normal sound data and the abnormal model expressing the abnormal sound data, and uses the minimum value as the threshold value. decide. The weight update unit (14) uses a plurality of normal sound data, abnormal sound data, and a threshold so that all the abnormal sound data are determined to be abnormal, and the probability that the normal sound data is determined to be abnormal is minimized. , Update the weight of the anomalous model.
展开▼