Novel Deep Learning Approaches for Single-Channel Speech Enhancement