def mlf2PhonemesAndTsList(inputFileName):
'''
parse output of alignment in mlf format ( with words)
output: phonemes with begin and end ts
# TODO: change automatically extension from txt to mlf
'''
allLines = loadTextFile(inputFileName)
listPhonemesAndTs = []
prevStartTime = -1
# when reading lines from MLF, skip first 2 and last
for line in allLines[2:-1]:
tokens = line.split(" ")
startTime = float(tokens[0]) / 10000000
endTime = float(tokens[1]) / 10000000
# if Praat does not allow insertion of new token with same timestamp. This happend when prev. token was 'sp'. So remove it and insert current
if (prevStartTime == startTime):
listPhonemesAndTs.pop()
phoneme = tokens[2].strip()
listPhonemesAndTs.append([startTime, endTime, phoneme])
# remember startTime
prevStartTime = startTime
return listPhonemesAndTs
def _loadSyllables(self, pathToSymbTrFile):
allLines = loadTextFile(pathToSymbTrFile)
currSyllable = None
currSyllTotalDuration = None
# skip first line.
for i in range( 1, len(allLines) ):
line = allLines[i]
line = line.replace(os.linesep,'') # remove end line in an os-independent way
line = line.replace('\r','')
tokens = line.split("\t")
if len(tokens) < 12:
print "TOKENS ARE 11, no syllable "; sys.exit()
# sanity check MINIMAL_DURATION of embelishments.
# hack: change crazy small notes to min duration. it does not matter a lot
if tokens[7] > MINIMAL_DURATION_UNIT and tokens[1] == '8':
tokens[7] = MINIMAL_DURATION_UNIT
currDuration = float(tokens[6]) / float(tokens[7]) * MINIMAL_DURATION_UNIT
currSyllable, currSyllTotalDuration = self.parseCurrTxtToken(currSyllable, currSyllTotalDuration, tokens, currDuration)
#end parsing loop
# store last
currSyllable.setDuration(currSyllTotalDuration)
self.listSyllables.append(currSyllable)
def _loadSectionBoundaries(self, URISectionFIle):
if not os.path.isfile(URISectionFIle):
sys.exit("no file {}".format(URISectionFIle))
ext = os.path.splitext(os.path.basename(URISectionFIle))[1]
if ext == '.tsv':
allLines = loadTextFile(URISectionFIle)
for line in allLines[1:]:
# triples of sectin name, start note number, end note number
tokens = line.strip().split("\t")
if not len(tokens) == 3:
sys.exit(
"tokens in line {} from file {} are not 3. make sure /t are used"
.format(line, URISectionFIle))
tmpTriplet = tokens[0], int(tokens[1]), int(tokens[2])
self.sectionboundaries.append(tmpTriplet)
######################
elif ext == '.json':
b = open(URISectionFIle)
scoreAnno = json.load(b)
b.close()
scoreSectionAnnos = scoreAnno['sections']
for section in scoreSectionAnnos:
tmpTriplet = section['name'], int(section['startNote']), int(
section['endNote'])
self.sectionboundaries.append(tmpTriplet)
def _loadSyllables(self, pathToSymbTrFile):
allLines = loadTextFile(pathToSymbTrFile)
# skip first line
for line in allLines[1:]:
line = line.replace(os.linesep,
'') # remove end line in a os-independent way
line = line.replace('\r', '')
tokens = line.split("\t")
if len(tokens) == 12:
# TUK ZABIVA.
if tokens[11] != '.' and tokens[11] != '. ' and tokens[
11] != '. ' and tokens[11] != 'SAZ' and tokens[
11] != 'SAZ ' and tokens[11] != 'SAZ ' and tokens[
11] != u'ARANA\u011eME' and tokens[
11] != u'ARANA\u011eME ' and tokens[
11] != u'ARANA\u011eME ' and tokens[
11] != u'ARANA\\\\u011eME' and tokens[
11] != u'ARANAGME':
# note number and syllable
text = tokens[11].replace('_', ' ')
tupleSyllable = int(tokens[0]), text
self.listSyllables.append(tupleSyllable)
def readLookupTable(withSynthesis):
'''
read lookup table from file
'''
if not withSynthesis:
lookupTableURI = os.path.join(
os.path.dirname(os.path.realpath(__file__)),
'grapheme2METUphonemeLookupTable')
else:
lookupTableURI = os.path.join(
os.path.dirname(os.path.realpath(__file__)),
'grapheme2METUphonemeLookupTableSYNTH')
lookupTableRaw = loadTextFile(lookupTableURI)
lookupTable = dict()
for lineTable in lookupTableRaw:
tokens = lineTable.split("\t")
grapheme = tokens[0]
if not isinstance(grapheme, unicode):
grapheme = unicode(grapheme, 'utf-8')
if len(grapheme) == 4 and grapheme[0].isdigit():
grapheme = "\u" + grapheme
grapheme = grapheme.decode('unicode-escape')
lookupTable[grapheme] = tokens[1].rstrip()
return lookupTable
def _loadsectionTimeStamps(self, URILinkedSectionsFile):
if not os.path.isfile(URILinkedSectionsFile):
sys.exit("no file {}".format(URILinkedSectionsFile))
ext = os.path.splitext(os.path.basename(URILinkedSectionsFile))[1]
if ext == '.txt' or ext == '.tsv':
lines = loadTextFile(URILinkedSectionsFile)
for line in lines:
tokens = line.split()
self.beginTs.append(float(tokens[0]))
self.endTs.append(float(tokens[1]))
self.sectionNamesSequence.append(tokens[2])
# WORKAROUND for section mapping. read mapping index from 4th field in .annotations file
# sanity check:
# if len(tokens) == 4:
# self.sectionIndices.append(int(tokens[3]))
#####################
elif ext == '.json':
b = open(URILinkedSectionsFile)
sectionLinks = json.load(b)
b.close()
sectionAnnos = sectionLinks['annotations']
for sectionAnno in sectionAnnos:
beginTimeStr = str(sectionAnno['time'][0])
beginTimeStr = beginTimeStr.replace("[", "")
beginTimeStr = beginTimeStr.replace("]", "")
endTimeStr = str(sectionAnno['time'][1])
endTimeStr = endTimeStr.replace("[", "")
endTimeStr = endTimeStr.replace("]", "")
self.beginTs.append(float(beginTimeStr))
self.endTs.append(float(endTimeStr))
self.sectionNamesSequence.append(str(sectionAnno['name']))
else:
sys.exit("section annotation file {} has not know file extension.".
format(URILinkedSectionsFile))
# match automatically section names from sectionLinks to scoreSections
indices = []
s1 = []
for s in self.makamScore.sectionToLyricsMap:
s1.append(s[0])
self.sectionIndices = matchSections(s1, self.sectionNamesSequence,
indices)
if len(self.sectionIndices) != len(self.sectionNamesSequence):
sys.exit(
"number of sections and number of matched sections not same!")
def readLookupTable(withSynthesis):
'''
read lookup table from file
'''
if not withSynthesis:
lookupTableURI= os.path.join(os.path.dirname(os.path.realpath(__file__)) , 'grapheme2METUphonemeLookupTable' )
else:
lookupTableURI= os.path.join(os.path.dirname(os.path.realpath(__file__)) , 'grapheme2METUphonemeLookupTableSYNTH' )
lookupTableRaw = loadTextFile(lookupTableURI)
lookupTable = dict()
for lineTable in lookupTableRaw:
tokens = lineTable.split("\t")
grapheme = tokens[0]
if not isinstance(grapheme, unicode):
grapheme = unicode(grapheme,'utf-8')
if len(grapheme) == 4 and grapheme[0].isdigit():
grapheme = "\u" + grapheme
grapheme = grapheme.decode('unicode-escape')
lookupTable[grapheme] = tokens[1].rstrip()
return lookupTable
def _loadSyllables(self, pathToSymbTrFile):
allLines = loadTextFile(pathToSymbTrFile)
currSyllable = None
currSyllTotalDuration = None
# skip first line.
for i in range(1, len(allLines)):
line = allLines[i]
line = line.replace(os.linesep,
'') # remove end line in an os-independent way
line = line.replace('\r', '')
tokens = line.split("\t")
if len(tokens) < 12:
print "TOKENS ARE 11, no syllable "
sys.exit()
# sanity check MINIMAL_DURATION of embelishments.
# hack: change crazy small notes to min duration. it does not matter a lot
if tokens[7] > MINIMAL_DURATION_UNIT and tokens[1] == '8':
tokens[7] = MINIMAL_DURATION_UNIT
currDuration = float(tokens[6]) / float(
tokens[7]) * MINIMAL_DURATION_UNIT
currSyllable, currSyllTotalDuration = self.parseCurrTxtToken(
currSyllable, currSyllTotalDuration, tokens, currDuration)
#end parsing loop
# store last
currSyllable.setDuration(currSyllTotalDuration)
self.listSyllables.append(currSyllable)
def mlf2WordAndTsList(inputFileName):
'''
parse output of alignment in mlf format ( with words)
output: words with begin and end ts
NOTE: length of tokens=5 if no -o option is set on HVite
TODO: change automatically extension from txt to mlf
'''
extracedWordList = []
LENGTH_TOKENS_NEW_WORD = 5
allLines = loadTextFile(inputFileName)
listWordsAndTs = allLines[2:-1]
currentTokenIndex = 0
tokens = listWordsAndTs[currentTokenIndex].split(" ")
while currentTokenIndex < len(listWordsAndTs):
# get begin ts
startTime = float(tokens[0]) / 10000000
wordMETU = tokens[-1].strip()
# move to next
prevTokens = tokens
currentTokenIndex += 1
# sanity check
if currentTokenIndex >= len(listWordsAndTs):
endTime = float(prevTokens[1]) / 10000000
extracedWordList.append([startTime, endTime, wordMETU])
break
tokens = listWordsAndTs[currentTokenIndex].split(" ")
# fast forward phonemes while end of word
while len(tokens
) == LENGTH_TOKENS_NEW_WORD - 1 and currentTokenIndex < len(
listWordsAndTs):
# end of word is last phoneme before 'sp'
if tokens[2] == "sp":
# move to next
currentTokenIndex += 1
if currentTokenIndex < len(listWordsAndTs):
tokens = listWordsAndTs[currentTokenIndex].split(" ")
break
prevTokens = tokens
currentTokenIndex += 1
tokens = listWordsAndTs[currentTokenIndex].split(" ")
# end of word. after inner while loop
endTime = float(prevTokens[1]) / 10000000
extracedWordList.append([startTime, endTime, wordMETU])
return extracedWordList