Code for "Context Aware Novelty Detection autoEncoder" by Ellen Rushe and Brian Mac Namee
Note for the Mimii Dataset: The same test data configuration was used as [1] with all novel sounds being used in the test set along with an equal number of normal segments. This left the remainder of the data to be used for training. A portion of normal examples from the test set was used for validation. Please note that these normal validation examples were also used in the evaluations for all models. This portion’s size was equal to 10% of the number of files in the training data (which are all normal files).
[1] Purohit, H., Tanabe, R., Ichige, K., Endo, T., Nikaido, Y., Suefusa, K., Kawaguchi, Y., 2019. Mimii dataset: Sound dataset for malfunctioning industrial machine investi- gation and inspection. arXiv preprint arXiv:1909.09347 .