So much developer rage in that Status :) On the HMM stuff, it was pretty much the baseline code from the link. The things I recall experimenting were more about getting it done faster (threading some training phases, different gcc options during synthesis etc).