marytts.tools.newlanguage
Class LexiconCreator

java.lang.Object
  marytts.tools.newlanguage.LexiconCreator

Direct Known Subclasses:

CMUDict2MaryFST

public class LexiconCreator
extends java.lang.Object

The LexiconCreator is the base class for creating the files needed to run the phonemiser component for a new language. From a list of phonetically transcribed words, the class will create:

• a lexicon file, efficiently stored as a Finite State Transducer;
• a letter-to-sound prediction file, as a decision tree in MARY format.

The input file is expected to contain data in the following format: grapheme | ' a l - l o - p h o n e s | (optional-part-of-speech) Hereby, the allophones must correspond to a defined allophone set, given in the constructor. The file's encoding is expected to be UTF-8. Subclasses of LexiconCreator can override prepareLexicon() to provide data in this format.

Author:

marc

See Also:

AllophoneSet

Field Summary

protected AllophoneSet	allophoneSet
protected int	context
protected boolean	convertToLowercase
protected java.lang.String	fstFilename
protected java.lang.String	lexiconFilename
protected org.apache.log4j.Logger	logger
protected java.lang.String	ltsFilename
protected boolean	predictStress

Constructor Summary

LexiconCreator(AllophoneSet allophoneSet, java.lang.String lexiconFilename, java.lang.String fstFilename, java.lang.String ltsFilename)
Initialise a new lexicon creator.

LexiconCreator(AllophoneSet allophoneSet, java.lang.String lexiconFilename, java.lang.String fstFilename, java.lang.String ltsFilename, boolean convertToLowercase, boolean predictStress, int context)
Initialise a new lexicon creator.

Method Summary

protected void	compileFST()
protected void	compileLTS()
void	createLexicon()
static void	main(java.lang.String[] args)
protected void	prepareLexicon() This base implementation does nothing.
protected void	testFST()
protected void	testLTS()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

logger

protected org.apache.log4j.Logger logger

allophoneSet

protected AllophoneSet allophoneSet

lexiconFilename

protected java.lang.String lexiconFilename

fstFilename

protected java.lang.String fstFilename

ltsFilename

protected java.lang.String ltsFilename

convertToLowercase

protected boolean convertToLowercase

predictStress

protected boolean predictStress

context

protected int context

Constructor Detail

LexiconCreator

public LexiconCreator(AllophoneSet allophoneSet,
                      java.lang.String lexiconFilename,
                      java.lang.String fstFilename,
                      java.lang.String ltsFilename)

Initialise a new lexicon creator. Letter to sound rules built with this lexicon creator will convert graphemes to lowercase before prediction, using the locale given in the allophone set; letter-to-sound rules will also predict stress; a context of 2 characters to the left and to the right of the current character will be used as predictive features.

Parameters:

allophoneSet - this specifies the set of phonetic symbols that can be used in the lexicon, and provides the locale of the lexicon

lexiconFilename - where to find the plain-text lexicon

fstFilename - where to create the compressed lexicon FST file

ltsFilename - where to create the letter-to-sound prediction tree.

LexiconCreator

public LexiconCreator(AllophoneSet allophoneSet,
                      java.lang.String lexiconFilename,
                      java.lang.String fstFilename,
                      java.lang.String ltsFilename,
                      boolean convertToLowercase,
                      boolean predictStress,
                      int context)

Initialise a new lexicon creator.

Parameters:

allophoneSet - this specifies the set of phonetic symbols that can be used in the lexicon, and provides the locale of the lexicon

lexiconFilename - where to find the plain-text lexicon

fstFilename - where to create the compressed lexicon FST file

ltsFilename - where to create the letter-to-sound prediction tree.

convertToLowercase - if true, Letter to sound rules built with this lexicon creator will convert graphemes to lowercase before prediction, using the locale given in the allophone set.

predictStress - if true, letter-to-sound rules will predict stress.

context - the number of characters to the left and to the right of the current character will be used as predictive features.

Method Detail

prepareLexicon

protected void prepareLexicon()
                       throws java.io.IOException

This base implementation does nothing. Subclasses can override this method to prepare a lexicon in the expected format, which should then be found at lexiconFilename.

Throws:

java.io.IOException

compileFST

protected void compileFST()
                   throws java.io.IOException

Throws:

java.io.IOException

testFST

protected void testFST()
                throws java.io.IOException

Throws:

java.io.IOException

compileLTS

protected void compileLTS()
                   throws java.io.IOException

Throws:

java.io.IOException

testLTS

protected void testLTS()
                throws java.io.IOException,
                       MaryConfigurationException

Throws:

java.io.IOException

MaryConfigurationException

createLexicon

public void createLexicon()
                   throws java.lang.Exception

Throws:

java.lang.Exception

main

public static void main(java.lang.String[] args)
                 throws java.lang.Exception

Parameters:

args -

Throws:

java.lang.Exception