netdict.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                  */ 
%/*                                                             */
%/*                                                             */
%/* ----------------------------------------------------------- */
%/* developed at:                                               */
%/*                                                             */
%/*      Speech Vision and Robotics group                       */
%/*      Cambridge University Engineering Department            */
%/*      http://svr-www.eng.cam.ac.uk/                          */
%/*                                                             */
%/*      Entropic Cambridge Research Laboratory                 */
%/*      (now part of Microsoft)                                */
%/*                                                             */
%/* ----------------------------------------------------------- */
%/*         Copyright: Microsoft Corporation                    */
%/*          1995-2000 Redmond, Washington USA                  */
%/*                    http://www.microsoft.com                 */
%/*                                                             */
%/*          2001-2002 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 - Steve Young 24/11/97
%

\mychap{Networks, Dictionaries and Language Models}{netdict}

\sidepic{Tool.netdict}{80}{ 
The preceding chapters have described how to process speech
data and how to train various types of HMM.
This and the following chapter are concerned with building
a speech recogniser using \HTK.  This chapter focuses on
the use of networks\index{networks} and dictionaries\index{dictionaries}.  
A network describes the
sequence of words that can be recognised and, for the case of sub-word
systems, a dictionary describes the sequence of HMMs that constitute
each word.
A word level network will typically represent either
a \textit{Task Grammar} which defines all of the legal word 
sequences explicitly
or a \textit{Word Loop} which simply puts all words of the vocabulary
in a loop and therefore allows any word to follow any other word.
Word-loop networks are often augmented by a stochastic language model.  
Networks can also be used
to define phone recognisers and various types of word-spotting systems.
}

Networks are specified using the \HTK\ \textit{Standard Lattice Format} (SLF)
which is described in detail in Chapter~\ref{c:htkslf}.
This is a general purpose text format which is used for representing
multiple hypotheses in a recogniser output as well as word networks.  
Since SLF\index{SLF} format is text-based, it can be written directly using any text editor.
However, this can be rather tedious and \HTK\ provides
two tools which allow the application designer to use a higher-level
representation.  Firstly, the tool \htool{HParse} allows networks
to be generated from a source text containing extended BNF format
grammar rules.  This format was the only grammar definition
language provided in earlier versions of \HTK\ and hence 
\htool{HParse} also provides backwards compatibility. 
\index{standard lattice format}

\htool{HParse} task grammars are very easy to write, but they 
do not allow fine control
over the actual network used by the recogniser. 
The tool \htool{HBuild} works directly at the SLF level to provide
this detailed control.  Its main function is to 
enable a large word network to be decomposed into
a set of small self-contained sub-networks using as input an extended
SLF format.  This enhances the
design process and avoids the need for unnecessary repetition.

\htool{HBuild} can also be used to perform a number
of special-purpose functions.  Firstly, it can construct 
word-loop and word-pair grammars automatically.  Secondly,
it can incorporate a statistical bigram
language model into a network.  These can be generated from label
transcriptions using \htool{HLStats}.  However,
\HTK\ supports the standard ARPA MIT-LL text format for backed-off
N-gram language models, and hence, import from other sources is possible.
 
Whichever tool is used to generate a word network, it is important
to ensure that the generated network represents the intended grammar.
It is also helpful to have some measure of the difficulty of the
recognition task.  To assist with this, the tool \htool{HSGen} is
provided.  This tool will generate example word sequences from
an SLF network using random sampling.  It will also estimate the
perplexity of the network.

When a word network is loaded into a recogniser, 
a dictionary is consulted to convert each
word in the network into a sequence of phone HMMs.   The dictionary can
have multiple pronunciations in which case several sequences may be joined
in parallel to make a word.  Options exist in this process to automatically
convert the dictionary entries to context-dependent triphone
models, either within a word or cross-word.  Pronouncing 
dictionaries are a vital resource in building speech recognition
systems and, in practice, word pronunciations can be derived from
many different sources.  The \HTK\ tool \htool{HDMan} enables a dictionary
to be constructed automatically from different sources.  Each source
can be individually edited and translated and merged to form a
uniform \HTK\ format dictionary.

The various facilities for describing a word network and expanding into a
HMM level network suitable for building a recogniser are implemented
by the \HTK\ library module \htool{HNet}.  The facilities for loading
and manipulating dictionaries are implemented by the \HTK\ library module
\htool{HDict} and  for loading
and manipulating language models are implemented by
\htool{HLM}.  These facilities and those provided by 
\htool{HParse}, \htool{HBuild}, \htool{HSGen}, 
\htool{HLStats} and \htool{HDMan} are
the subject of this chapter.

\mysect{How Networks are Used}{netuse}

Before delving into the details of word networks\index{networks!in recognition} and dictionaries, it will
be helpful to understand their r\^{o}le in building a speech recogniser
using \HTK.  Fig~\href{f:recsys} illustrates the overall recognition
process.  A word network is defined using HTK Standard Lattice Format
(SLF).  An SLF word network is just a text file and it can be written
directly with a text editor or a tool can be used to build it. \HTK\ provides 
two such tools, \htool{HBuild} and
\htool{HParse}.  These both take as input a textual description and
output an SLF file. 
% Another way to generate SLF files is to use Entropic's \textit{grapHvite} package, which includes a 
% graphical tool that allows the required networks to be constructed on
% the screen.  
Whatever method is chosen, word network SLF generation 
is done \textit{off-line}
and is part of the system build process.

An SLF file contains a list of nodes representing words and a
list of arcs representing the transitions between words.   The transitions
can have probabilities attached to them and these can be used to indicate
\textit{preferences} in a grammar network.  They can also be used to
represent bigram probabilities in a back-off bigram network and 
\htool{HBuild} can generate such a bigram network automatically.
In addition to an SLF file, a \HTK\ recogniser requires a 
dictionary to supply pronunciations for each word in the network
and a set of acoustic HMM phone models. 
Dictionaries are input via the \HTK\ interface module \htool{HDict}.

The dictionary, HMM set and word network are input to the \HTK\ library
module \htool{HNet} whose function is to generate an equivalent network of
HMMs. Each word in the dictionary may have several pronunciations and in
this case there will be one branch in the network corresponding to each
alternative pronunciation. Each pronunciation may consist either of a list
of phones or a list of HMM names. In the former case, \htool{HNet} can
optionally expand the HMM network to use either word internal triphones or
cross-word triphones.
Once the HMM network has been constructed, it can be
input to the decoder module \htool{HRec} and used to recognise
speech input.  Note that HMM network construction is performed \textit{on-line}
at recognition time as part of the initialisation process.

\centrefig{recsys}{100}{Overview of the Recognition Process}

For convenience, \HTK\ provides a recognition\index{recognition!overall process} tool called \htool{HVite}
to allow the functions provided by \htool{HNet} and \htool{HRec}
to be invoked from the command line. \htool{HVite} is particularly
useful for running experimental evaluations on test speech stored
in disk files and for basic testing using live audio input.
However, application developers
should note that \htool{HVite} is just a shell containing calls to
load the word network, dictionary and models; generate the recognition
network and then repeatedly recognise each input utterance.
For embedded applications, it may well be appropriate to
dispense with \htool{HVite} and call the functions in 
\htool{HNet} and \htool{HRec} directly from the application.
The use of \htool{HVite} is explained in the next chapter.

\mysect{Word Networks and Standard Lattice Format}{slfintro}

\index{standard lattice format}
This section provides a basic introduction to the \HTK\ Standard Lattice
Format (SLF). SLF files are used for a variety of functions some of
which lie beyond the scope of the standard \HTK\ package.   The
description here is limited to those features of SLF which are required
to describe word networks suitable for input to \htool{HNet}.  The
following Chapter describes the further features of SLF used for
representing the output of a recogniser. For reference, a full
description of SLF is given in Chapter~\ref{c:htkslf}.

\index{SLF!format}
A word network in SLF\index{SLF} consists of a list of nodes and a list of arcs.  
The nodes represent words and the arcs represent the transition between
words\footnote{More precisely, nodes represent the ends of
words and arcs represent the transitions between word ends.
This distinction becomes important when describing
recognition output since acoustic scores are attached
to arcs not nodes. }.  
Each node and arc definition is written on a single line and
consists of a number of fields. Each field specification consists of a
``name=value'' pair. Field names can be any length but all commonly used
field names consist of a single letter.  By convention, field names
starting  with a capital letter are mandatory  whereas field names
starting with a lower-case letter are optional.  Any line beginning with
a \texttt{\#} is a comment and is ignored.

\centrefig{wdnet}{80}{A Simple Word Network}

The following example should illustrate the basic format \index{SLF!word network}
of an SLF word network file.  It corresponds to the network
illustrated in Fig~\href{f:wdnet} which represents all sequences
consisting of the words ``bit'' and ``but'' starting with the
word ``start'' and ending with the word ``end''.  As will be
seen later, the start and end words will be mapped to a silence
model so this grammar allows speakers to 
say ``bit but but bit bit ....etc''.
\begin{verbatim}
    # Define size of network: N=num nodes and L=num arcs
    N=4 L=8
    # List nodes: I=node-number, W=word
    I=0 W=start
    I=1 W=end
    I=2 W=bit
    I=3 W=but
    # List arcs: J=arc-number, S=start-node, E=end-node
    J=0 S=0 E=2
    J=1 S=0 E=3
    J=2 S=3 E=1
    J=3 S=2 E=1
    J=4 S=2 E=3
    J=5 S=3 E=3
    J=6 S=3 E=2
    J=7 S=2 E=2
\end{verbatim}
Notice that the first line which defines the size of the network must be
given before any node or arc definitions.
A node is a \textit{network start node} if it has no predecessors,
and a node is \textit{network end node} if it has no successors.
There must be one and only one network start node and one network
end node.
In the above, node 0 is a network start node and node 1 is a
network end node.
The choice of the names ``start'' and ``end'' for these nodes
has no significance.

\centrefig{wdnet1}{80}{A Word Network Using Null Nodes}

A word network can have null nodes indicated by the special
predefined word name \texttt{!NULL}.  Null nodes are useful for reducing
the number of arcs required.  For example, the \textit{Bit-But}
network could be defined as follows\index{SLF!null nodes}
\begin{verbatim}
    # Network using null nodes
    N=6 L=7
    I=0 W=start
    I=1 W=end
    I=2 W=bit
    I=3 W=but
    I=4 W=!NULL
    I=5 W=!NULL
    J=0 S=0 E=4
    J=1 S=4 E=2
    J=2 S=4 E=3
    J=3 S=2 E=5
    J=4 S=3 E=5
    J=5 S=5 E=4
    J=6 S=5 E=1
\end{verbatim}
In this case, there is no significant saving, however, if there
were many words in parallel, the total number of arcs would be
much reduced by using null nodes to form common start and end points for
the loop-back connections.

By default, all arcs are equally likely.  However, the optional
field \texttt{l=x} can be used to attach the log transition probability
\texttt{x} to an arc.  For example, if the word ``but'' was twice
as likely as ``bit'', the arcs numbered 1 and 2 in the last example
could be changed to
\begin{verbatim}
    J=1 S=4 E=2 l=-1.1