Training acoustic model for CMU Sphinx using Finite state grammar (JSGF)

Question

I need to train my own acoustic model. On the official wiki, this process is described for ngramm models in DMP format, but I need to use a (grammar) language model provided in JSGF format. Is there a such possibility? And if so, could you give a tutorial on this subject?

Strictly

Answer 1

Modifications are minimial. Open psdecode.pl script in the installed location or in scripts and change

-lm => $ST::DEC_CFG_LANGUAGEMODEL,

to

-jsgf => "/full/path/to/jsgf/file",

Syst

Thanks! but.. can i set this thing anywhere in config files? I'm working on adapting simon for using sphinx, and on the basis of this problem, I can not afford to change the script on the user's machine. – Strictly

No, it's not possible to do it in the configuration files. If you want to suggest some new feature to be included into sphinxtrain, you are welcome to do so sourceforge.net/tracker/?group_id=1904&atid=301904 – Syst