hlrescore.tex

该压缩包为最新版htk的源代码,htk是现在比较流行的语音处理软件,请有兴趣的朋友下载使用

%/* ----------------------------------------------------------- */
%/*                                                             */
%/*                          ___                                */
%/*                       |_| | |_/   SPEECH                    */
%/*                       | | | | \   RECOGNITION               */
%/*                       =========   SOFTWARE                  */ 
%/*                                                             */
%/*                                                             */
%/* ----------------------------------------------------------- */
%/*         Copyright:                                          */
%/*              2002  Cambridge University                     */
%/*                    Engineering Department                   */
%/*                                                             */
%/*   Use of this software is governed by a License Agreement   */
%/*    ** See the file License for the Conditions of Use  **    */
%/*    **     This banner notice must not be removed      **    */
%/*                                                             */
%/* ----------------------------------------------------------- */
%
% HTKBook - Gunnar Evermann 10.12.2002
%

\newpage
\mysect{HLRescore}{HLRescore}

\mysubsect{Function}{HLRescore-Function}

\index{hlrescore@\htool{HLRescore}|(}
\htool{HLRescore} is a general lattice post-processing tool. It reads
lattices (for example produced by \htool{HVite}) and applies one of
the following operations on them:

\begin{itemize}
\item finding 1-best path through lattice
\item pruning lattice using forward-bacward scores
\item expanding lattices with new language model
\item calculating various lattice statistics
\end{itemize}

A typical scenario for the use of \htool{HLRescore} is the application
of a higher order n-gram to the word lattices generated with HVite and
a bigram. This would involve the following steps:

\begin{itemize}
\item lattice generation with HVite using a bigram
\item lattice pruning with HLRescore (\texttt{-t})
\item expansion of lattices using a trigram (\texttt{-n})
\item finding 1-best transcription in the expanded lattice
  (\texttt{-f})
\end{itemize}

Another common use of HLRescore is the tuning of the language
model scaling factor and the word insertion penalty for use in
recognition. Instead of having to re-run a decoder many times with
different parameter settings the decoder is run once to generate
lattices. \htool{HLRescore} can be used to find the best transcription
for a give parameter setting very quickly. These different
transcriptions can then be scored (using \htool{HResults}) and the
parameter setting that yields the lowest word error rate can be
selected. 



\mysubsect{Use}{HLRescore-Use}

\htool{HLRescore} is invoked via the command line
\begin{verbatim}
   HLRescore [options] vocabFile LatFiles......
\end{verbatim}

\htool{HLRescore} reads each of the lattice files and performs te
requested operation(s) on them. At least one of the following
operations must be selected: find 1-best (\texttt{-f}), write lattices
(\texttt{-w}), calculate statistics (\texttt{-c}).


The detailed operation of \htool{HLRescore} is controlled by the following
command line options
\begin{optlist}

  \ttitem{-i mlf} Output transcriptions to master file \texttt{mlf}.

  \ttitem{-l s} Directory in which to store label/lattice files.

  \ttitem{-n lm} Load ARPA-format n-gram language model from file
  \texttt{lm} and expand lattice with this LM. All acoustic scores are
  unchanged but the LM scores are replaced and lattices nodes (i.e.\
  contexts) are expanded as required by the structure of the LM.

  \ttitem{-o s} Choose how the output labels should be formatted.
        \texttt{s} is a string with certain letters (from \texttt{NSCTWM})
        indicating binary flags that control formatting options.
        \texttt{N} normalise acoustic scores by dividing by the duration
        (in frames) of the segment.
        \texttt{S} remove scores from output label.  By default 
        scores will be set to the total likelihood of the segment.
        \texttt{C} Set the transcription labels to start and end on
        frame centres. By default start times are set to the start
        time of the frame and end times are set to the end time of 
        the frame.
        \texttt{T} Do not include times in output label files.
        \texttt{W} Do not include words in output label files
        when performing state or model alignment.
        \texttt{M} Do not include model names in output label
        files.

  \ttitem{-t f [a]} Perform lattice pruning after reading lattices with
  beamwidth \texttt{f}. If second argument is given lower beam to
  limit arcs per second to \texttt{a}.

  \ttitem{-u f} Perform lattice pruning before writing output
  lattices. Otherwise like \texttt{-t}.

  \ttitem{-p f} Set the word insertion log probability to \texttt{f} 
  (default 0.0).
  
  \ttitem{-a f} Set the acoustic model scale factor to \texttt{f}.
  (default value 1.0).

  \ttitem{-r f} Set the dictionary pronunciation probability scale 
        factor to \texttt{f}. (default value 1.0).

  \ttitem{-s f} Set the grammar scale factor to \texttt{f}.
        This factor post-multiplies the language model likelihoods
        from the word lattices.  (default value 1.0).

  \ttitem{-d} Take pronunciation probabilities from the dictionary
  instead of from the lattice.

  \ttitem{-c} Calculate and output lattice statistics.

  \ttitem{-f} Find 1-best transcription (path) in lattice.

  \ttitem{-w} Write output lattice after processing.

  \ttitem{-q s} Choose how the output lattice should be formatted.
         \texttt{s} is a string with certain letters (from \texttt{ABtvaldmn})
         indicating binary flags that control formatting options.
         \texttt{A} attach word labels to arcs rather than nodes.
         \texttt{B} output lattices in binary for speed.
         \texttt{t} output node times.
         \texttt{v} output pronunciation information.
         \texttt{a} output acoustic likelihoods.
         \texttt{l} output language model likelihoods.
         \texttt{d} output word alignments (if available).
         \texttt{m} output within word alignment durations.
         \texttt{n} output within word alignment likelihoods.

  \ttitem{-y ext}  This sets the extension for output label files to
        \texttt{ext} (default \texttt{rec}).

\stdoptF
\stdoptG
\stdoptH
\stdoptI
\stdoptJ
\stdoptK
\stdoptP

\end{optlist}
\stdopts{HLRescore}

\mysubsect{Tracing}{HLRescore-Tracing}

\htool{HLRescore} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}
   \ttitem{0001} enable basic progress reporting.  
   \ttitem{0002} output generated transcriptions.
   \ttitem{0004} show details of lattice I/O
   \ttitem{0010} show memory usage after each lattice
\end{optlist}
Trace flags are set using the \texttt{-T} option or the \texttt{TRACE} 
configuration variable.
\index{hlrescore@\htool{HLRescore}|)}


%%% Local Variables: 
%%% mode: latex
%%% TeX-master: "../htkbook"
%%% End: