hsgen.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  31/10/95
%

\newpage
\mysect{HSGen}{HSGen}

\mysubsect{Function}{HSGen-Function}

\index{hsgen@\htool{HSGen}|(}
This program will read in a word network definition in standard 
\HTK\ lattice format representing a Regular Grammar $G$ and randomly
generate sentences from the language $L(G)$ of $G$.  The sentences are
written to standard output, one per line and an option is provided
to number them if required.

The empirical entropy $H_e$ can also be calculated using the formula
\begin{equation}
  H_e = \frac{\sum_k P(S_k)}{\sum_k |S_k|}
\end{equation}
where $S_k$ is the $k$'th sentence generated and $|S_k|$ is its length.
The perplexity $P_e$ is computed from $H_e$ by
\begin{equation}
  P_e = 2^{H_e}
\end{equation}
The probability of each sentence $P(S_k)$ is computed from the product of
the individual branch probabilities.


\mysubsect{Use}{HSGen-Use}

\htool{HSGen} is invoked by the command line
\begin{verbatim}
   HSGen [options] wdnet dictfile
\end{verbatim}
where \texttt{dictfile} is a dictionary containing all of the words
used in the word network stored in \texttt{wdnet}.  This dictionary
is only used as a word list, the pronunciations are ignored.

The available options are

\begin{optlist}

  \ttitem{-l} When this option is set, each generated sentence 
              is preceded by a line number.

  \ttitem{-n N}  This sets the total number of sentences generated
              to be \texttt{N} (default value 100).

  \ttitem{-q}  Set quiet mode.  This suppresses the printing of sentences.
         It is useful when estimating the entropy of $L(G)$ since the
    accuracy of the latter depends on the number of sentences generated.

  \ttitem{-s}  Compute word network statistics.  When set, the number of
    network nodes, the vocabulary size, the empirical entropy, the
    perplexity, the average sentence length, the minimum sentence length
    and the maximum sentence length 
    are computed and printed on the standard output.
\end{optlist}
\stdopts{HSLab}

\mysubsect{Tracing}{HSGen-Tracing}

\htool{HSLab} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}
   \ttitem{00001} basic progress reporting 
   \ttitem{00002} detailed trace of lattice traversal
\end{optlist}
Trace flags are set using the \texttt{-T} option or the  \texttt{TRACE} 
configuration variable.
\index{hsgen@\htool{HSGen}|)}


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