After a quick look, it seems Julius doesn't use the new deep-learning stuff? In terms of data, http://www.openslr.org/12/ http://www.openslr.org/12/ says it h

After a quick look, it seems Julius doesn't use the new deep-learning stuff? In terms of data, http://www.openslr.org/12/ http://www.openslr.org/12/ says it has 300 hours + of speech+text from librivox audiobooks. Using Librovox recordings seemed a great idea for making a freely available large dataset.