hlmcopy.tex

隐马尔可夫模型源代码

%
% HTKBook - Julian Odell    14/08/97
%
% Updated - Gareth Moore    15/01/02
%

\newpage
\mysect{HLMCopy}{HLMCopy}

\mysubsect{Function}{HLMCopy-Function}

\index{hlmcopy@\htool{HLMCopy}|(}
The basic function of this tool is to copy language models. During this
operation the target model can be optionally adjusted to a specific vocabulary,
reduced in size by applying pruning parameters to the different $n$-gram
components and written out in a different file format. Previously unseen words
can be added to the language model with unigram entries supplied in a unigram
probability file. 
At the same time, the tool can be used to extract word pronunciations from a
number of source dictionaries and output a target dictionary for a specified word
list. \htool{HLMCopy} is a key utility enabling the user to construct custom 
dictionaries and language models tailored to a particular recognition task.

\mysubsect{Use}{HLMCopy-Use}

\htool{HLMCopy} is invoked by the command line
\begin{verbatim}
   HLMCopy [options] inLMFile outLMFile
\end{verbatim}
This copies the language model {\tt inLMFile} to {\tt outLMFile} optionally
applying editing operations controlled by the following options.
\begin{optlist}

  \ttitem{-c n c} Set the pruning threshold for $n$-grams to $c$. 
	Pruning can be applied to the bigram and higher
	components of a model ($n$>1). The pruning procedure will keep only 
	$n$-grams which have been observed more than $c$ times. Note
	that this option is only applicable to count-based language 
        models.

  \ttitem{-d f} Use dictionary {\tt f} as a source of pronunciations
        for the output dictionary. A set of dictionaries can be
        specified, in order of priority, with multiple {\tt -d}
        options.
  
  \ttitem{-f s} Set the output language model format to {\tt s}.
        Possible options are {\tt TEXT} for the standard ARPA-MIT
	LM format, {\tt BIN} for Entropic {\em binary} format and 
        {\tt ULTRA} for Entropic {\em ultra} format.
        
  \ttitem{-n n} Save target model as $n$-gram.

  \ttitem{-m} Allow multiple identical pronunciations for a single
	word.  Normally identical pronunciations are deleted.  This
	option may be required when a single word/pronunciation has
	several different output symbols.

  \ttitem{-o} Allow pronunciations for a single word to be selected
	from multiple dictionaries.  Normally the dictionaries are
	prioritised by the order they appear on the command line with
	only entries in the first dictionary containing a
	pronunciation for a particular word being copied to the output
	dictionary.

  \ttitem{-u f} Use unigrams from file {\tt f} as replacements for the
	ones in the language model itself.  Any words appearing in the
	output language model which have entries in the unigram file
	(which is formatted as {\tt LOG10PROB WORD}) use the
	likelihood ({\tt log10(prob)}) from the unigram file rather
	than from the language model.  This allows simple language
	model adaptation as well as allowing unigram probabilities to
	be assigned words in the output vocabulary that do not appear
	in the input language model.  In some instances you may wish
	to use \htool{LNorm} to renormalise the model after using {\tt
	-u}.

  \ttitem{-v f}  Write a dictionary covering the output vocabulary to
	file {\tt f}.  If any required words cannot be found in the
	set of input dictionaries an error will be generated.

  \ttitem{-w f} Read a word-list defining the output vocabulary from
	{\tt f}. This will be used to select the vocabulary for both
	the output language model and output dictionary.

\end{optlist}
\stdopts{HLMCopy}

\mysubsect{Tracing}{HLMCopy-Tracing}

\htool{HLMCopy} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}
   \ttitem{00001} basic progress reporting.
\end{optlist}
Trace flags are set using the \texttt{-T} option or the  \texttt{TRACE} 
configuration variable.
\index{hlmcopy@\htool{HLMCopy}|)}