Spatial Diffuseness Features for DNN-Based Speech Recognition in Noisy and Reverberant Environments