Abstract: Speech enhancement and dereverberation approaches based on neural networks are designed to learn a transformation from noisy to clean speech using supervised learning. However, networks ...