Fig. 3From: Phone recognition with hierarchical convolutional deep maxout networksPhone error rate as a function of the input context size. The context size is given in terms of framesBack to article page