Heiga ZEN (Byung Ha CHUN) wrote:
Hi, Thomas Drugman wrote (2007/10/26 18:59):I would like to train my HMMs by imposing them a given decision tree.I suppose changes have to be made in the editfiles called by HHEd during tree-based context clustering steps ($CXCL1->4). I guess trees should be loaded and then states should be clustered according to them, but I cannot fix how (how to force LT command to recognize leaves?).HTK/HTS do not support such operation.
There might be a workaround, although I haven't actually tried this. HHEd will not impose a decision tree on a set of models, but it will *synthesise* unseen models using a tree. 'unseen models' are merely ones that do not exist in the MMF and model list initially loaded by HHEd
So, how about this:- you have a decision tree previously learned; in the tree, there is a macro name at each leaf
- load an MMF into HHEd which contains only the definitions of these macros, but no actual HMMs (i.e. you have deleted the HMM definitions from the MMF, manually or using a script); I guess you also have to give it an empty model list file on the command line
- use HHEd's "AU" command to synthesise *all* the models you need (the ones you previously deleted from the MMF, plus any unseen ones)
- the models are now defined in terms of the macros, just like any tied-parameter model set
- save the MMFThe only possible problem I see is that you are loading macros into HHEd that are not used (at the time they are loaded - because there are no models), so you need to make sure it doesn't simply delete unused macros during loading. There is a function in HHEd.c called 'PurgeMacros' but I think it is only called just before saving the MMF, so that's OK.
Simon