hled.tex

隐马尔可夫模型源代码

%/* ----------------------------------------------------------- */
%/*                                                             */
%/*                          ___                                */
%/*                       |_| | |_/   SPEECH                    */
%/*                       | | | | \   RECOGNITION               */
%/*                       =========   SOFTWARE                  */ 
%/*                                                             */
%/*                                                             */
%/* ----------------------------------------------------------- */
%/*         Copyright: Microsoft Corporation                    */
%/*          1995-2000 Redmond, Washington USA                  */
%/*                    http://www.microsoft.com                */
%/*                                                             */
%/*   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 - Steve Young  1/12/97
%

\newpage
\mysect{HLEd}{HLEd}

\mysubsect{Function}{HLEd-Function}

\index{hled@\htool{HLEd}|(}
This program is a simple editor for manipulating label files.
Typical examples of its use might be to merge a sequence of
labels into a single composite label or to expand a set of
labels into a context sensitive set.  \htool{HLEd} works by
reading in a list of {\em editing} commands from an edit
script file and then makes an edited copy of one or more 
label files.  For multiple level files, edit commands are applied
to the \textit{current level} which is initially the first (i.e.\ 1).
Other levels may be edited by moving to the required level using the
\texttt{ML} Move Level command.

Each edit command in the script file must be
on a separate line.  The first two-letter mnemonic on each line is the command name and
the remaining letters denote labels\footnote{In earlier versions of 
HTK, \htool{HLEd} command names consisted of a single letter.  These
are still supported for backwards compatibility and they are included
in the command summary produced using the \texttt{-Q} option.  
However, commands
introduced since version 2.0  have two letter names.}.
The commands supported may be divided into
two sets.  Those in the first set are used to edit individual
labels and they are as follows

\begin{varlist}
   \fwitem{2cm}{CH X A Y B}      Change \texttt{Y} in the context of \texttt{A\_B} 
      to \texttt{X}.  \texttt{A} and/or \texttt{B} may be a \texttt{*} to match 
      any context, otherwise they must be defined by a \texttt{DC} command
      (see below).  A
      block of consecutive \texttt{CH} commands are effectively executed in
      parallel so that the contexts are those that exist before any of
      the commands in the block are applied.
   \fwitem{2cm}{DC A B C ..}     define the context \texttt{A} as the set of labels
        \texttt{B}, \texttt{C}, etc.   
   \fwitem{2cm}{DE A B ..}     Delete any occurrences of labels \texttt{A}
        or \texttt{B} etc. 
   \fwitem{2cm}{FI A Y B}  Find \texttt{Y} in the context of \texttt{A\_B} 
      and count the number of occurrences.
   \fwitem{2cm}{ME X A B ..}     Merge any sequence of labels \texttt{A}
       \texttt{B} \texttt{C} etc.\  and call the new segment \texttt{X}.
   \fwitem{2cm}{ML N} Move to label level \texttt{N}.
   \fwitem{2cm}{RE X A B ..}   Replace all occurrences of labels \texttt{A}
      or \texttt{B} etc.\ by the label \texttt{X}.
\end{varlist}

The commands in the second set perform global operations on whole
transcriptions.  They are as follows.

\begin{varlist}
   \fwitem{2cm}{DL [N]} Delete all labels in the current level.  If the
     optional integer arg is given, then level \texttt{N} is deleted.
   \fwitem{2cm}{EX} Expand all labels either from words to phones using the 
        first pronunciation from a dictionary when it is specified on the 
        command line otherwise expand labels of the form 
        \texttt{A\_B\_C\_D\_...} into a sequence of separate labels 
        \texttt{A B C D ....}.  This is useful for label formats which 
        include a complete orthography as a single label or for creating 
        a set of sub-word labels from a word orthography for a sub-word 
        based recogniser.  When a label is expanded in this way, the 
        label duration is divided into equal length segments.  This can
        only be performed on the root level of a multiple level file.
   \fwitem{2cm}{FG X} Mark all unlabelled segments of the input
       file of duration greater than $T_g$ msecs with the label \texttt{X}.
       The default value for $T_g$ is 50000.0 (=5msecs) but this
       can be changed using the \texttt{-g} command line option.  This
       command
       is mainly used for explicitly labelling inter-word silences in
       data files for which only the actual speech has been transcribed.
   \fwitem{2cm}{IS A B}  Insert label \texttt{A} at the start of every
       transcription and \texttt{B} at the end.  This command is usually
       used to insert silence labels.
   \fwitem{2cm}{IT} Ignore triphone contexts in \texttt{CH} and \texttt{FI}
       commands.
   \fwitem{2cm}{LC [X]} Convert all phoneme labels to left context
       dependent.  If \texttt{X} is given then the first phoneme label
       \texttt{a} becomes \texttt{X-a} otherwise it is left unchanged.
   \fwitem{2cm}{NB X} The label  \texttt{X} (typically a short pause)
       should be ignored at word boundaries when using the context
       commands \texttt{LC}, \texttt{RC} and \texttt{TC}.
   \fwitem{2cm}{RC [X]} Convert all phoneme labels to right context
       dependent.  If \texttt{X} is given then the last phoneme label
       \texttt{z} becomes \texttt{z+X} otherwise it is left unchanged.
   \fwitem{2cm}{SB X}  Define the label \texttt{X} to be a sentence boundary
       marker.  This label can then be used in context-sensitive change
       commands.
   \fwitem{2cm}{SO}     Sort all labels into time order.
   \fwitem{2cm}{SP}     Split multiple levels into multiple alternative label lists.
   \fwitem{2cm}{TC [X[Y]]} Convert all phoneme labels to Triphones, that is
       left and right context
       dependent.  If \texttt{X} is given then the first phoneme label
       \texttt{a} becomes \texttt{X-a+b} otherwise it is left unchanged.
       If \texttt{Y} is given then the last phoneme label
       \texttt{z} becomes \texttt{y-z+Y} otherwise  if \texttt{X} is given then it
        becomes \texttt{y-z+X} otherwise it is left unchanged.
   \fwitem{2cm}{WB X}  Define \texttt{X} to be an inter-word label. 
       This command affects the operation of the \texttt{LC}, \texttt{RC}
       and \texttt{TC} commands.  The expansion of context labels is 
       blocked wherever an inter-word label occurs.
\end{varlist}

The source and target label file formats can be 
defined using the \texttt{-G} and \texttt{-P} command line arguments.  They can
also be set using the configuration variables \texttt{SOURCELABEL} 
and \texttt{TARGETLABEL}.  The default for both cases is the HTK format.

\mysubsect{Use}{HLEd-Use}

\htool{HLEd} is invoked by typing the command line
\begin{verbatim}
   HLEd [options] edCmdFile labFiles ..
\end{verbatim}
This causes \htool{HLEd} to be applied to each \texttt{labFile} in turn
using the edit commands listed in \texttt{edCmdFile}.  The 
\texttt{labFiles} may be master label files.
The available options are

\begin{optlist}

  \ttitem{-b} Suppress label boundary times in output files.

  \ttitem{-d s} Read a dictionary from file {\tt s} and use this
  for expanding labels when the {\tt EX} command is used.

  \ttitem{-i mlf} This specifies that the output transcriptions
  are written to the master label file \texttt{mlf}.

  \ttitem{-g t} Set the minimum gap detected by the \texttt{FG} to be \texttt{t}
      (default 50000.0 = 5msecs).
     All gaps of shorter duration than \texttt{t} are ignored and not labelled.
  
  \ttitem{-l s} Directory to store output label files (default is current directory).
      When output is directed to an MLF,  this option can be used to
      add a path to each output file name.  In particular, setting the option
      \verb+-l '*'+ will cause a label file named \texttt{xxx} to be prefixed
      by the pattern \verb+"*/xxx"+ in the output MLF file.  This is useful
      for generating MLFs which are independent of the location of the 
      corresponding data files.

  \ttitem{-m} Strip all labels to monophones on loading.
 
  \ttitem{-n fn} This option causes a list of all new label names
      created to be output to the file \texttt{fn}.

\stdoptG
\stdoptI
\stdoptP
\stdoptX

\end{optlist}
\stdopts{HLEd}

\mysubsect{Tracing}{HLEd-Tracing}

\htool{HLEd} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}
   \ttitem{000001} basic progress reporting.
   \ttitem{000002} edit script details.
   \ttitem{000004} general command operation.
   \ttitem{000010} change operations.
   \ttitem{000020} level split/merge operations.
   \ttitem{000040} delete level operation.
   \ttitem{000100} edit file input.
   \ttitem{000200} memory usage.
   \ttitem{000400} dictionary expansion in \texttt{EX} command
\end{optlist}
Trace flags are set using the \texttt{-T} option or the  \texttt{TRACE} 
configuration variable.
\index{hled@\htool{HLEd}|)}


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