Provides an implementation of the LanguageModel interface.
This implementation loads large binary forms of language model
generated by the lm_convert CMU Sphinx tool.
To handle n-gram language models with n>3, use the lm_convert tool
available in the cmusphinx/branches/LIUM/lm_tools_32bits_file64bits
directory of the CMU Sphinx svn repository on SourceForge.
The DMP binary tools are built by lm_convert from ARPA language models
produced by, for example, the CMU-Cambridge Statistical Language Modeling
Toolkit or the SRILM toolkit. Be careful to sort all the n-grams in the same
way, for each n, and to have always a value for back-off weights (even 0).