g2p.py

#!/usr/bin/env python

"""
Grapheme-to-Phoneme Conversion

Samples can be either in plain format (one word per line followed by
phonetic transcription) or Bliss XML Lexicon format.
"""

__author__    = 'Maximilian Bisani'
__version__   = '$LastChangedRevision: 95 $'
__date__      = '$LastChangedDate: 2007-06-02 16:32:35 +0200 (Sat, 02 Jun 2007) $'
__copyright__ = 'Copyright (c) 2004-2005  RWTH Aachen University'
__license__   = """
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License Version 2 (June
1991) as published by the Free Software Foundation.
 
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, you will find it at
http://www.gnu.org/licenses/gpl.html, or write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110,
USA.
 
Should a provision of no. 9 and 10 of the GNU General Public License
be invalid or become invalid, a valid provision is deemed to have been
agreed upon which comes closest to what the parties intended
commercially. In any case guarantee/warranty shall be limited to gross
negligent actions or intended actions or fraudulent concealment.
"""

import math, sets, sys
import SequiturTool
from sequitur import Translator
from misc import gOpenIn, gOpenOut, set

# ===========================================================================
def loadPlainSample(fname, encoding = None):
    sample = []
    for line in gOpenIn(fname, encoding or defaultEncoding):
        fields = line.split()
        if not fields: continue
        left  = tuple(fields[0])
        right = tuple(fields[1:])
        sample.append((left, right))
    return sample

def pronunciationsFromXmlLexicon(xml):
    pronunciations = {}
    lexicon = xml.getroot()
    for lemma in lexicon.getiterator('lemma'):
        orth = [ orth.text.strip() for orth in lemma.findall('orth') if orth.text is not None ]
        phon = [ tuple((phon.text or '').split()) for phon in lemma.findall('phon') ]
        for w in orth:
            if w in pronunciations:
                pronunciations[w] += phon
            else:
                pronunciations[w] = phon
    return pronunciations

def loadBlissLexicon(fname):
    from elementtree.ElementTree import ElementTree
    xml = ElementTree(file = gOpenIn(fname))
    pronunciations = pronunciationsFromXmlLexicon(xml)
    result = [ (orth, phon)
               for orth in pronunciations
               if not (orth.startswith('[') and orth.endswith(']'))
               for phon in pronunciations[orth] ]
    result.sort()
    return result

def loadG2PSample(fname):
    if fname == '-':
        sample = loadPlainSample(fname)
    else:
        firstLine = gOpenIn(fname, defaultEncoding).readline()
        if firstLine.startswith('<?xml'):
            sample = [ (tuple(orth), tuple(phon))
                       for orth, phon in loadBlissLexicon(fname) ]
        else:
            sample = loadPlainSample(fname)
    return sample

def loadP2PSample(compfname):
    fnames = compfname.split(':')
    assert len(fnames) == 2
    left  = dict(loadG2PSample(fnames[0]))
    right = dict(loadG2PSample(fnames[1]))
    sample = []
    for w in set(left.keys()) & set(right.keys()):
        sample.append((left[w], right[w]))
    return sample

# ===========================================================================
def readApply(fname):
    for line in gOpenIn(fname, defaultEncoding):
        word = line.strip()
        left = tuple(word)
        yield word, left

def readApplyP2P(fname):
    for line in gOpenIn(fname, defaultEncoding):
        fields = line.split()
        word = fields[0]
        left = tuple(fields[1:])
        yield word, left

# ===========================================================================
class MemoryTranslator:
    def __init__(self, sample):
        self.memory = dict(sample)

    TranslationFailure = Translator.TranslationFailure

    def __call__(self, left):
        if left in self.memory:
            return self.memory[left]
        else:
            raise self.TranslationFailure()

    def reportStats(self, f):
        pass

# ===========================================================================
def mainTest(translator, testSample, options):
    if options.shouldTranspose:
        testSample = SequiturTool.transposeSample(testSample)
    if options.testResult:
        resultFile = gOpenOut(options.testResult, defaultEncoding)
    else:
        resultFile = None
    from Evaluation import Evaluator
    evaluator = Evaluator()
    evaluator.setSample(testSample)
    evaluator.resultFile = resultFile
    evaluator.verboseLog = stdout
    if options.test_segmental:
        supraSegmental = set(['.', "'", '"'])
        def removeSupraSegmental(phon):
            return filter(lambda p: p not in supraSegmental, phon)
        evaluator.compareFilter = removeSupraSegmental
    result = evaluator.evaluate(translator)
    print >> stdout, result

def mainApply(translator, options):
    if options.phoneme_to_phoneme:
        words = readApplyP2P(options.applySample)
    elif options.shouldTranspose:
        words = readApplyP2P(options.applySample)
    else:
        words = readApply(options.applySample)

    if options.variants_mass or options.variants_number:
        wantVariants = True
        threshold = options.variants_mass or 1.0
        nVariantsLimit = options.variants_number or 1e9
    else:
        wantVariants = False

    for word, left in words:
        try:
            if wantVariants:
                totalPosterior = 0.0
                nVariants = 0
                nBest = translator.nBestInit(left)
                while totalPosterior < threshold and nVariants < nVariantsLimit:
                    try:
                        logLik, result = translator.nBestNext(nBest)
                    except StopIteration:
                        break
                    posterior = math.exp(logLik - nBest.logLikTotal)
                    print >> stdout, ('%s\t%d\t%f\t%s' % \
                           (word, nVariants, posterior, ' '.join(result)))
                    totalPosterior += posterior
                    nVariants += 1
            else:
                result = translator(left)
                print >> stdout, ('%s\t%s' % (word, ' '.join(result)))
        except translator.TranslationFailure, exc:
            try:
                print >> stderr, 'failed to convert "%s": %s' % (word, exc)
            except Exception:
                pass


def main(options, args):
    if options.phoneme_to_phoneme:
        loadSample = loadP2PSample
    else:
        loadSample = loadG2PSample

    if options.fakeTranslator:
        translator = MemoryTranslator(loadSample(options.fakeTranslator))
    else:
        model = SequiturTool.procureModel(options, loadSample, log=stdout)
        if not model:
            return 1
        if options.testSample or options.applySample:
            translator = Translator(model)
            if options.stack_limit:
                translator.setStackLimit(options.stack_limit)
        del model

    if options.testSample:
        mainTest(translator, loadSample(options.testSample), options)
        translator.reportStats(sys.stdout)

    if options.applySample:
        mainApply(translator, options)
        translator.reportStats(sys.stderr)

# ===========================================================================
if __name__ == '__main__':
    import optparse, tool
    optparser = optparse.OptionParser(
        usage   = '%prog [OPTION]... FILE...\n' + __doc__,
        version = '%prog ' + __version__)
    tool.addOptions(optparser)
    SequiturTool.addTrainOptions(optparser)
    optparser.add_option(
        '-e', '--encoding', default='ISO-8859-15',
        help='use character set encoding ENC', metavar='ENC')
    optparser.add_option(
        '-P', '--phoneme-to-phoneme', action='store_true',
        help='train/apply a phoneme-to-phoneme converter')
    optparser.add_option(
        '--test-segmental', action='store_true',
        help='evaluate only at segmental level, i.e. do not count syllable boundaries and stress marks')
    optparser.add_option(
        '-B', '--result', dest='testResult',
        help='store test result in table FILE (for use with bootlog or R)', metavar='FILE')
    optparser.add_option(
        '-a', '--apply', dest='applySample',
        help='apply grapheme-to-phoneme conversion to words read from FILE', metavar='FILE')
    optparser.add_option(
        '-V', '--variants-mass', type='float',
        help='generate pronunciation variants until \sum_i p(var_i) >= Q (only effective with --apply)', metavar='Q')
    optparser.add_option(
        '--variants-number', type='int',
        help='generate up to N pronunciation variants (only effective with --apply)', metavar='N')
    optparser.add_option(
        '-f', '--fake', dest='fakeTranslator',
        help='use a translation memory (read from sample FILE) instead of a genuine model (use in combination with -x to evaluate two files against each other)', metavar='FILE')
    optparser.add_option(
        '--stack-limit', type='int',
        help='limit size of search stack to N elements', metavar='N')

    options, args = optparser.parse_args()

    import codecs
    global defaultEncoding
    defaultEncoding = options.encoding
    global stdout, stderr
    encoder, decoder, streamReader, streamWriter = codecs.lookup(options.encoding)
    stdout = streamWriter(sys.stdout)
    stderr = streamWriter(sys.stderr)

    tool.run(main, options, args)