hinit.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{HInit}{HInit}

\mysubsect{Function}{HInit-Function}

\index{hinit@\htool{HInit}|(}
\htool{HInit} is used to provide initial estimates for the parameters
of a single HMM using a set of observation sequences.
It works by repeatedly using Viterbi alignment to segment the
training observations and then recomputing the parameters
by pooling the vectors in each segment.   For mixture Gaussians, each
vector in each segment is aligned with the component with the highest
likelihood.   Each cluster of vectors then determines the parameters
of the associated mixture component.
In the absence of an initial model, the process
is started by performing a uniform 
segmentation of each training observation and for mixture Gaussians,
the vectors in each uniform segment are clustered using a modified K-Means 
algorithm\footnote{This algorithm is significantly different from 
earlier versions of HTK where K-means clustering was used at every
iteration and the Viterbi alignment was limited to states}. 

\htool{HInit} can be used to provide initial estimates of whole word models
in which case the observation sequences are realisations of the 
corresponding vocabulary word.
Alternatively, \htool{HInit} can be used to generate initial estimates of
{\em seed} HMMs for phoneme-based speech recognition.
In this latter case, the observation sequences will consist
of segments of continuously spoken training material.  \htool{HInit} will
{\it cut} these out of the training data automatically by simply
giving it a segment label.

In both of the above applications, \htool{HInit} normally takes 
as input a prototype
HMM definition which defines the required HMM topology i.e.\ it has
the form of the required HMM except that means, variances and mixture
weights are ignored\footnote{Prototypes should either
have GConst set (the value does
not matter) to avoid {\tt HTK} trying to compute it or
variances should be set to a positive value such as 1.0 to
ensure that GConst is computable}.  The
transition matrix of the prototype specifies both the allowed
transitions and their initial probabilities.  Transitions which
are assigned zero probability will remain zero and hence denote
non-allowed transitions.  \htool{HInit} estimates transition probabilities
by counting the number of times each state is visited during
the alignment process.

\htool{HInit} supports multiple mixtures, multiple streams,
parameter tying within a single model, full or diagonal
covariance matrices, tied-mixture models and discrete models.  
The output of HInit is typically input to HRest.

Like all re-estimation tools, \htool{HInit} allows a floor to  be set on
each individual variance by defining a variance floor macro for each
data stream (see chapter~\ref{c:Training}).
 
\mysubsect{Use}{HInit-Use}

\htool{HInit} is invoked via the command line
\begin{verbatim}
   HInit [options] hmm trainFiles ...
\end{verbatim}
This causes the means and variances of the given {\tt hmm} to be
estimated repeatedly using the data in {\tt trainFiles} 
until either a maximum iteration limit is reached
or the estimation converges.
The HMM definition can be contained within one or more macro
files loaded via the standard \texttt{-H} option.  Otherwise, the 
definition will be read from a file called \texttt{hmm}.
The list of train files can be stored in a script file if required.

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

  \ttitem{-e f} This sets the convergence factor to the real value {\tt f}.
      The convergence factor is the relative change between successive
      values of $P_{max}({O}|\lambda)$ computed as a by-product
      of the Viterbi alignment stage (default value 0.0001).

  \ttitem{-i N} This sets the maximum number of estimation cycles
      to {\tt N} (default value 20).

  \ttitem{-l s} The string {\tt s} must be the name of a
      segment label.  When this option is used, \htool{HInit} searches
      through all of the training files and cuts out all segments with
      the given label.  When this option is not used, \htool{HInit} assumes that
      each training file is a single token.

  \ttitem{-m N}  This sets the minimum number of training examples so
      that if fewer than {\tt N} examples are supplied an error is
      reported (default value 3).

  \ttitem{-n}  This flag suppresses the initial uniform
      segmentation performed by \htool{HInit} allowing it to be used
      to update the parameters of an existing model.
  
  \ttitem{-o s}  The string {\tt s} is used as the name of the output
      HMM in place of the source name.  This is provided in \htool{HInit}
      since it is often used to initialise a model from a prototype
      input definition.  The default is to use the source name.

  \ttitem{-u flags}  By default, \htool{HInit} updates all 
      of the HMM parameters,
      that is, means, variances, mixture weights
      and transition probabilities. This 
      option causes just the parameters indicated by the {\tt flags}
      argument to be updated, this argument is a string containing one
      or more of the letters {\tt m} (mean), {\tt v} (variance), 
      {\tt t} (transition) and {\tt w} (mixture weight).  The presence of a
      letter enables the updating of the corresponding parameter set.

  \ttitem{-v f}  This sets the minimum variance (i.e. diagonal element of
      the covariance matrix) to the real value {\tt f} The default value
      is 0.0.

  \ttitem{-w f}  Any mixture weight  or discrete observation probability
      which falls below the global
            constant {\tt MINMIX} is treated as being zero.
      When this parameter is  set,  all mixture weights  are floored
      to {\tt f * MINMIX}.

\stdoptB
\stdoptF
\stdoptG
\stdoptH
\stdoptI
\stdoptL
\stdoptM
\stdoptX

\end{optlist}
\stdopts{HInit}

\mysubsect{Tracing}{HInit-Tracing}

\htool{HInit} supports the following trace options where each
trace flag is given using an octal base
\begin{optlist}
   \ttitem{000001} basic progress reporting.
   \ttitem{000002} file loading information.
   \ttitem{000004} segments within each file.
   \ttitem{000010} uniform segmentation.
   \ttitem{000020} Viterbi alignment.
   \ttitem{000040} state alignment.
   \ttitem{000100} mixture component alignment.
   \ttitem{000200} count updating.
   \ttitem{000400} output probabilities.
\end{optlist}
Trace flags are set using the \texttt{-T} option or the  \texttt{TRACE} 
configuration variable.
\index{hinit@\htool{HInit}|)}


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