Supervised model training for overlapping sound events based on unsupervised source separation | IEEE Conference Publication | IEEE Xplore