An Experimental Analysis of Deep Learning Architectures for Supervised Speech Enhancement