|
|
Initialize filler probabilities (penalties, whatever) module and return a pointer to the structure created. Filler word probabilities are simple unigram probabilities. Here is an example of such a file (one entry per line; a word and a probability): <sil> 0.10792 <uh> 0.00866 <um> 0.00147 If the first character in a line is a '#', the line is treated as a comment and ignored. If no filler probabilities file is provided, the silence word gets silprob, and all other filler words get fillprob. As with the trigram LM, the resulting log-probabilities are multiplied by a language weight and finally a word insertion penalty is tacked on. - Parameters:
-
| dict |
In: Dictionary containing filler words |
| file |
In: Filler word probabilities file, if any |
| silprob |
In: Default probability for silence word |
| fillprob |
In: Default probability for non-silence filler words |
| lw |
In: Language weight (see lm.h) |
| wip |
In: Word insertion penalty (see lm.h) |
|
