def getNbVowels(_utt):
i = 0
for phone in _utt.phoneList:
if utils.isVowel(phone).find('-1') != -1 or \
utils.isVowel(phone).find('0') != -1 or \
phone.end == _utt.end:
pass
else:
i = i + 1
return i
def extractVowelDur(_textgridFilename):
"""
This function extracts the durations of the vowels in a word
Input
_textgridFilename filename of the textgrid to read
Output
vowelDur list of viwel durations in a word
"""
currUtterance = utils.parseTextgrid(_textgridFilename)
currPhoneList = currUtterance.phoneList
vowelDur = []
for phonei in currPhoneList:
if utils.isVowel(phonei).find('-1') != -1 or \
utils.isVowel(phonei).find('0') != -1 or \
phonei.end == currUtterance.end:
pass
else:
dur = float(phonei.end) - float(phonei.start)
vowelDur.append(dur)
return vowelDur
def getConsecutiveVowelRatio(_textgridFilename):
"""
This function extracts the ratio between each consecutive pair of vowel
read from the textGrid file given in input
"""
vowelRatio = []
vowelDur = []
currUtterance = utils.parseTextgrid(_textgridFilename)
currPhoneList = currUtterance.phoneList
for phonei in currPhoneList:
if utils.isVowel(phonei).find('-1') != -1 or \
utils.isVowel(phonei).find('0') != -1 or \
phonei.end == currUtterance.end:
pass
else:
dur = float(phonei.end) - float(phonei.start)
vowelDur.append(dur)
for i in range(len(vowelDur) - 1):
ratio = vowelDur[i + 1] / vowelDur[i]
vowelRatio.append(ratio)
return vowelRatio
def getDurationVec(_textgridFilename):
"""
This function returns the list of 25 durations according to the
phone nature of the current sample point (vowel, consonant, insertion)
Input
_textgridFilename filename of the textgrid file to parse
Output
durations list of 25 durations
"""
listOfNatures = [] # Size should be 25
durations = [] # Size should be 25
currentUtterance = utils.parseTextgrid(_textgridFilename)
# Time step
nbPoints = 25
timeStep = (float(currentUtterance.end) -
float(currentUtterance.start)) / nbPoints
currentPhoneList = currentUtterance.phoneList
phoneIndex = 0
for i in range(1, nbPoints + 1):
mid = (timeStep * i - timeStep * (i - 1)) / 2
if (float(currentPhoneList[phoneIndex].end) -
float(currentPhoneList[0].start) > (timeStep * (i - 1) + mid)):
listOfNatures.append(utils.isVowel(currentPhoneList[phoneIndex]))
if utils.isVowel(currentPhoneList[phoneIndex]).find('0') != -1 or \
utils.isVowel(currentPhoneList[phoneIndex]).find('-1') != -1 or \
phoneIndex == (len(currentPhoneList) -1):
durations.append(0)
else:
dur = float(currentPhoneList[phoneIndex].end) - float(
currentPhoneList[phoneIndex].start)
durations.append(dur)
else:
phoneIndex += 1
listOfNatures.append(utils.isVowel(currentPhoneList[phoneIndex]))
if utils.isVowel(currentPhoneList[phoneIndex]).find('0') != -1 or\
utils.isVowel(currentPhoneList[phoneIndex]).find('-1') != -1 or\
phoneIndex == (len(currentPhoneList) -1):
durations.append(0)
else:
dur = float(currentPhoneList[phoneIndex].end) - float(
currentPhoneList[phoneIndex].start)
durations.append(dur)
return durations
def getArticleFeatures(parsedSents, sentences):
errors = []
features = []
headList = []
for i in range(len(parsedSents)):
if parsedSents[i] == None:
continue
s = parsedSents[i]
sent = sentences[i]
sHeads = set()
for chunk in s.chunks:
if chunk.type == 'NP':
#filter NP chunk size bigger than 4
if len(chunk.words) > 4:
continue
head = utils.getHeadFeatures(chunk)
if head == None or head.type == 'PRP':
continue #ignore PRP and None
#get NP article
article, s_article = utils.getArticle(chunk)
if s_article == 'O':
continue #ignore DT which is not a, an or the
#precheck a/an singular and plural of this NP
c_list = utils.getCountable(head)
isUncount = False
if c_list[0] == 'N' and c_list[1] == 'Y':
isUncount = True
error = utils.precheckArticle(article, sent, head, chunk,
isUncount)
if error != None:
e = dict()
e['start'] = error.start
e['end'] = error.end
e['output'] = error.output
e['desc'] = error.description
e['type'] = error.type_
e['original'] = error.original
e['newSent'] = error.newSent
errors.append(e)
continue
#for the rest of the NP chunk, extract features for classifier
bef3pos, aft3pos = utils.getAround(head.index, s, article)
bnp_word, bnp_pos = utils.getBeforeNP(chunk, s)
anp_word, anp_pos = utils.getAfterNP(chunk, s)
wordnet = utils.getWordNet(head)
isPlural = utils.getCount(head)
pp, pps = utils.prepModify(chunk)
adj, adj_grade, pdt, prp, relation = utils.getChunkFeatures(
chunk)
ref = utils.getRef(headList, head, sHeads)
vowel = utils.isVowel(chunk, article)
#features
feature = []
feature.append(s_article)
feature.append(head.string)
feature.append(head.lemma)
feature.append(head.type)
feature.append(isPlural)
feature.append(wordnet)
feature.append(pp)
feature.append(pps)
feature.append(adj)
feature.append(adj_grade)
feature.append(pdt)
feature.append(prp)
feature.append(relation)
feature.append(ref)
feature.append(bnp_word)
feature.append(bnp_pos)
feature.append(anp_word)
feature.append(anp_pos)
feature.extend(c_list)
feature.extend(bef3pos)
feature.extend(aft3pos)
#index of NP chunk, head and article
index = []
index.append(i)
index.append(chunk.start)
index.append(chunk.stop)
index.append(head.index)
if article == None:
#if need article where to insert?
index.append(-1)
else:
index.append(article.index)
index.append(vowel)
r = dict()
r["feature"] = feature
r["index"] = index
features.append(r)
sHeads.add(head.string.lower())
headList.append(sHeads)
if len(headList) > 5:
del headList[0]
return errors, features