Model converter needed

[dhdaines]

The acoustic model format has diverged somewhat from CMU Sphinx, and is expected to diverge further. Supporting multiple model formats is not consistent with the goal of making the smallest possible library, so we require a converter to be able to use publically available models. Currently this means:

• Convert text to binary model definition
• Convert mixture_weights to sendump
• Rename text files to include ".txt" extension
• Convert feat_params to JSON
• Include default dictionary

In the future it may mean (but this is not in the scope of this issue):

• Dictionary is an FST and may be a G2P model
• Model definition is also an FST (i.e. the "HC" in "HCLG")
• GMMs are also quantized
• GMMs might be replaced by DNNs