Have tested the tool a little bit for audio, and see a potential here. Especially useful for anyone who has a relatively large amount of unlabeled data, and want to be efficient in terms of what samples to spend resources labeling.