marytts.tools.newlanguage

Class LTSTrainer

java.lang.Object

marytts.fst.AlignerTrainer

marytts.tools.newlanguage.LTSTrainer

public class LTSTrainer
extends AlignerTrainer

This class is a generic approach to predict a phone sequence from a grapheme sequence. the normal sequence of steps is: 1) initialize the trainer with a phone set and a locale 2) read in the lexicon, preserve stress if you like 3) make some alignment iterations (usually 5-10) 4) train the trees and save them in wagon format in a specified directory see main method for an example. Apply the model using TrainedLTS

Author:

benjaminroth

Field Summary

Fields

Modifier and Type	Field and Description
protected boolean	considerStress
protected int	context
protected boolean	convertToLowercase
protected AllophoneSet	phSet

Fields inherited from class marytts.fst.AlignerTrainer

graphemeSet, inSplit, logger, optInfo, outSplit

Constructor Summary

Constructors

Constructor and Description
LTSTrainer(AllophoneSet aPhSet, boolean convertToLowercase, boolean considerStress, int context) Create a new LTSTrainer.

Method Summary

Methods

Modifier and Type	Method and Description
static void	main(String[] args)
void	readLexicon(BufferedReader lexicon, String splitPattern) reads in a lexicon in text format, lines are of the kind: graphemechain | phonechain | otherinformation Stress is optionally preserved, marking the first vowel of a stressed syllable with "1".
void	readLexicon(HashMap<String,String> lexicon) reads in a lexicon in text format, lines are of the kind: graphemechain | phonechain | otherinformation Stress is optionally preserved, marking the first vowel of a stressed syllable with "1".
void	save(CART tree, String saveTreefile) Convenience method to save files to graph2phon.wagon and graph2phon.pfeats in a specified directory with UTF-8 encoding.
CART	trainTree(int minLeafData) Train the tree, using binary decision nodes.

Methods inherited from class marytts.fst.AlignerTrainer

addAlreadySplit, addAlreadySplit, addAlreadySplit, addAlreadySplit, align, alignIteration, getAlignment, getAlignmentString, getInfoAlignment, getInputSyms, lexiconSize, splitAndAdd

Methods inherited from class java.lang.Object

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

Field Detail

phSet

protected AllophoneSet phSet

context

protected int context

convertToLowercase

protected boolean convertToLowercase

considerStress

protected boolean considerStress

Constructor Detail

LTSTrainer

public LTSTrainer(AllophoneSet aPhSet,
          boolean convertToLowercase,
          boolean considerStress,
          int context)

Create a new LTSTrainer.

Parameters:

aPhSet - the allophone set to use.

convertToLowercase - whether to convert all graphemes to lowercase, using the locale of the allophone set.

considerStress - indicator if stress is preserved

context - context

Method Detail

trainTree

public CART trainTree(int minLeafData)
               throws IOException

Train the tree, using binary decision nodes.

Parameters:

minLeafData - the minimum number of instances that have to occur in at least two subsets induced by split

Returns:

bigTree

Throws:

IOException - IOException

save

public void save(CART tree,
        String saveTreefile)
          throws IOException

Convenience method to save files to graph2phon.wagon and graph2phon.pfeats in a specified directory with UTF-8 encoding.

Parameters:

tree - tree

saveTreefile - saveTreefile

Throws:

IOException - IOException

readLexicon

public void readLexicon(BufferedReader lexicon,
               String splitPattern)
                 throws IOException

reads in a lexicon in text format, lines are of the kind: graphemechain | phonechain | otherinformation Stress is optionally preserved, marking the first vowel of a stressed syllable with "1".

Overrides:

readLexicon in class AlignerTrainer

Parameters:

lexicon - reader with lines of lexicon

splitPattern - a regular expression used for identifying the field separator in each line.

Throws:

IOException - IOException

readLexicon

public void readLexicon(HashMap<String,String> lexicon)

reads in a lexicon in text format, lines are of the kind: graphemechain | phonechain | otherinformation Stress is optionally preserved, marking the first vowel of a stressed syllable with "1".

Parameters:

lexicon - lexicon

main

public static void main(String[] args)
                 throws IOException,
                        MaryConfigurationException

Throws:

IOException

MaryConfigurationException