hcopy.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  24/11/97
%

\newpage
\mysect{HCopy}{HCopy}

\subsection{Function}

\index{hcopy@\htool{HCopy}|(}
This program will copy one or more data files to a designated output
file, optionally converting the data into a parameterised form.  While
the source files can be in any supported format, the output format is
always \HTKFF.  By default, the whole of the source file is copied to
the target but options exist to only copy a specified segment.  Hence,
this program is used to convert data files in other formats to the \HTKFF\ 
format, to concatenate or segment data files, and to parameterise the
result.  If any option is set which leads to the extraction of a segment
of the source file rather than all of it, then segments will be
extracted from all source files and concatenated to the target.

Labels will be copied/concatenated if any of the options indicating
labels are specified (\texttt{-i -l -x -G -I -L -P -X}). In this case, each
source data file must have an associated label file, and a target label
file is created. The name of the target label file is the root name of
the target data file with the extension \texttt{.lab}, unless the \texttt{-X}
option is used.  This new label file will contain the appropriately
copied/truncated/concatenated labels to correspond with the target data
file; all start and end boundaries are recalculated if necessary. 

When used in conjunction with \htool{HSLab}, \htool{HCopy} provides a facility for tasks
such as cropping silence surrounding recorded utterances.  Since input
files may be coerced, \htool{HCopy} can also be used to convert the parameter
kind of a file, for example from WAVEFORM to MFCC, depending on the
configuration options. 
Not all possible conversions can actually be performed; see Table~\href{t:validcons} for a list of valid conversions. Conversions must be specified via a configuration file as described in chapter~\ref{c:speechio}. Note also that the parameterisation qualifier \texttt{\_N} cannot be used when saving files to disk, and is meant only for on-the-fly parameterisation. 

\subsection{Use}

\htool{HCopy} is invoked by typing the command line
\begin{verbatim}
   HCopy [options] sa1 [ + sa2 + ... ] ta  [ sb1 [ + sb2 + ... ] tb ... ]
\end{verbatim}
This causes the contents of the one or more source files 
\texttt{sa1}, \texttt{sa2}, \ldots
 to be concatenated
and the result copied to the given target file \texttt{ta}. To avoid the overhead
of reinvoking the tool when processing large databases, multiple
sources and targets may be specified, for example
\begin{verbatim}
HCopy srcA.wav + srcB.wav tgtAB.wav srcC.wav tgtD.wav
\end{verbatim}
will create two new files \texttt{tgtAB.wav} and  \texttt{tgtD.wav}.
\htool{HCopy} takes file arguments from a script specified using the \texttt{-S} option
exactly as from the
command line, except that any newlines are ignored. 

The allowable options to \htool{HCopy} are as follows where all times
and durations are given in 100 ns units and
are written as floating-point numbers.


\begin{optlist}

 \ttitem{-a i}  Use level i of associated label files with the \texttt{-n}
   and  \texttt{-x} options.  Note that this is not the same as using the
 \texttt{TRANSLEVEL} configuration variable since the  \texttt{-a} option
still allows all levels to be copied through to the output files.

  \ttitem{-e f} End copying from the source file at time \texttt{f}.  The
 default is the end of the file. If \texttt{f} is negative or zero, it is
 interpreted as a time relative to the end of the file, while a positive value
 indicates an absolute time from the start of the file.

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

 \ttitem{-l s} Output label files to the directory \texttt{s}.
    The default is to output to the current directory.
  
 \ttitem{-m t} Set a margin of duration \texttt{t} around the
    segments defined by the \texttt{-n} and \texttt{-x} options. 

 \ttitem{-n i [j]} Extract the speech segment corresponding to the {\tt
     i}'th label in the source file.  If \texttt{j} is specified, then the
     segment corresponding to the sequence of labels \texttt{i} to \texttt{j}
     is extracted. Labels are numbered from their position in the
     label file.  A negative index can be used to count from the end
    of the label list.   Thus, \texttt{-n 1 -1} would specify the segment
    starting at the first label and ending at the last.

  \ttitem{-s f} Start copying from the source file at time \texttt{f}. 
    The default is 0.0, ie the beginning of the file.

  \ttitem{-t n} Set the line width to \texttt{n} chars when formatting
    trace output.

  \ttitem{-x s [n]} Extract the speech segment corresponding to the
       first occurrence of label \texttt{s} in the source file.  If \texttt{n}
       is specified, then the \texttt{n}'th occurrence is extracted. If
       multiple files are being concatenated, segments are extracted
       from each file in turn, and the label must exist for each
       concatenated file.

\stdoptF
\stdoptG
\stdoptI
\stdoptL
\stdoptO
\stdoptP
\stdoptX

\end{optlist}
\stdopts{HCopy}

Note that the parameter kind conversion 
mechanisms described in chapter~\ref{c:speechio}
will be applied to all source files.  In particular, if an automatic
conversion is requested via the configuration file, then \htool{HCopy} will copy
or concatenate the converted source files, not the actual contents.
Similarly, automatic byte swapping may occur depending on the source
format and the configuration variable \texttt{BYTEORDER}. Because the
sampling rate may change during conversions, the options that 
specify a position within a file i.e. \texttt{-s} and \texttt{-e}
use absolute times rather than sample index numbers.  All times in \HTK\
are given in units of 100ns and
are written as floating-point numbers. To save writing long strings of zeros, 
standard 
exponential notation may be used, for example \texttt{-s 1E6} indicates a
start time of 0.1 seconds from the beginning of the file.

\begin{center}
\begin{tabular}{|r||ccccccccccc|} \cline{2-12}
 \multicolumn{1}{c}{} & \multicolumn{11}{|c|}{\it Outputs } \\
\cline{2-11} \cline{2-11} \hline
          ~ & ~ & ~ & ~ & L  & ~ & ~ & ~ & ~ & ~ & ~ & ~ \\ 
          ~ & W & ~ & ~ & P  & ~ & ~ & ~ & ~ & ~ & D & ~ \\ 
          ~ & A & ~ & ~ & C  & ~ & ~ & ~ & M & ~ & I & ~ \\ 
          ~ & V & ~ & L & E  & I & ~ & ~ & E & ~ & S & ~ \\ 
          ~ & E & ~ & P & P  & R & ~ & F & L & ~ & C & ~ \\ 
          ~ & F & ~ & R & S  & E & M & B & S & U & R & ~ \\ 
          ~ & O & L & E & T  & E & F & A & P & S & E & P \\ 
          ~ & R & P & F & R  & F & C & N & E & E & T  & L \\ 
{\it Inputs} & M & C & C & A  & C & C & K & C & R & E & P \\ \hline 
~WAVEFORM & $\surd$ & $\surd$ & $\surd$ & $\surd$  &$\surd$& $\surd$ & $\surd$ & $\surd$ & ~ &$\surd$ & ~\\ 
~~~~~~LPC & ~ & $\surd$ & $\surd$ & $\surd$  &$\surd$& ~ & ~ & ~ & ~ &$\surd$ & ~\\ 
~~~LPREFC & ~ & $\surd$ & $\surd$ & $\surd$  &$\surd$& ~ & ~ & ~ & ~ &$\surd$& ~\\
LPCEPSTRA & ~ & $\surd$ & $\surd$ & $\surd$  &$\surd$& ~ & ~ & ~ & ~ &$\surd$& ~\\
~~~~IREFC & ~ & $\surd$ & $\surd$ & $\surd$  &$\surd$& ~ & ~ & ~ & ~ &$\surd$& ~\\
~~~~~MFCC & ~ & ~ & ~ &  ~ & ~ & $\surd$ & ~ & ~ & ~ &$\surd$& ~\\ 
~~~~FBANK & ~ & ~ & ~ &  ~ & ~ & $\surd$ & $\surd$ & ~ & ~ &$\surd$& ~\\
~~MELSPEC & ~ & ~ & ~ &  ~ & ~ & $\surd$ & $\surd$ & $\surd$ & ~ &$\surd$& ~\\
~~~~~USER & ~ & ~ & ~ &  ~ & ~ & ~ & ~ & ~ & $\surd$ &$\surd$ & ~\\
~DISCRETE & ~ & ~ & ~ &  ~ & ~ & ~ & ~ & ~ & ~ & $\surd$ & ~ \\ 
~~~~~~PLP & ~ & ~ & ~ &  ~ & ~ & ~ & ~ & ~ & ~ & $\surd$ & $\surd$ \\ 
\hline
\end{tabular}
\tabcap{validcons}{Valid Parameter Conversions}
\end{center}

Note that truncations are performed {\em after\/} any desired coding,
which may result in a loss of time resolution if the target file format
has a lower sampling rate. Also, because of windowing effects,
truncation, coding, and concatenation operations are not necessarily
interchangeable.  If in doubt, perform all truncation/concatenation in
the waveform domain and then perform parameterisation as a last, separate
invocation of \htool{HCopy}.

\subsection{Trace Output}

\htool{HCopy} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}

\ttitem{00001}  basic progress reporting. 
\ttitem{00002}  source and target file formats and parameter kinds.
\ttitem{00004}  segment boundaries computed from label files.
\ttitem{00010}  display memory usage after processing each file.
\end{optlist}
Trace flags are set using the \texttt{-T} option or the  \texttt{TRACE} 
configuration variable.
\index{hcopy@\htool{HCopy}|)}


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