Attention and Localization based on a Deep Convolutional Recurrent Model for Weakly Supervised Audio Tagging
https://arxiv.org/abs/1703.06052