def lineWordCount(textgrid_file, score_file):
'''
:param textgrid_file: annotation file
:return: numLines, numWords, numDians
'''
numLines, numWords, numDians, numPhos = 0, 0, 0, 0
wordDurationList, dianDurationList, nvcDurationList, vcDurationList = [],[],[],[]
entireLine = textgridParser.textGrid2WordList(textgrid_file,
whichTier='line')
entireWordList = textgridParser.textGrid2WordList(textgrid_file,
whichTier='pinyin')
entireDianList = textgridParser.textGrid2WordList(textgrid_file,
whichTier='dian')
entirePhoList = textgridParser.textGrid2WordList(textgrid_file,
whichTier='details')
utterance_durations, bpm = scoreParser.csvDurationScoreParser(score_file)
# parser word list for each line,
if len(entireWordList):
nestedWordLists, numLines, _ = textgridParser.wordListsParseByLines(
entireLine, entireWordList)
nestedWordLists_filtered = []
numWords = 0
for ii, wordList in enumerate(nestedWordLists):
if int(bpm[ii]):
# omit non score lines
nestedWordLists_filtered.append(wordList)
numWords += len(wordList[1])
numLines = len(nestedWordLists_filtered)
wordDurationList = wordDuration(nestedWordLists_filtered)
if len(entireDianList):
nestedWordLists, _, _ = textgridParser.wordListsParseByLines(
entireLine, entireDianList)
nestedWordLists_filtered = []
numDians = 0
for ii, wordList in enumerate(nestedWordLists):
if int(bpm[ii]):
nestedWordLists_filtered.append(wordList)
numDians += len(wordList[1])
dianDurationList = wordDuration(nestedWordLists_filtered)
if len(entirePhoList):
nestedWordLists, _, _ = textgridParser.wordListsParseByLines(
entireLine, entirePhoList)
nestedWordLists_filtered = []
numPhos = 0
for ii, wordList in enumerate(nestedWordLists):
if int(bpm[ii]):
nestedWordLists_filtered.append(wordList)
numPhos += len(wordList[1])
nvcDurationList, vcDurationList = phoDuration(nestedWordLists_filtered)
return numLines, numWords, numDians, numPhos, wordDurationList, dianDurationList, nvcDurationList, vcDurationList
def getBoundaryNumber(recordings, dataset_path):
listOnset = []
for i_recording, recording_name in enumerate(recordings):
groundtruth_textgrid_file = os.path.join(aCapella_root, dataset_path, annotation_path, recording_name+'.TextGrid')
score_file = os.path.join(aCapella_root, dataset_path, score_path, recording_name+'.csv')
if not os.path.isfile(score_file):
print 'Score not found: ' + score_file
continue
lineList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='line')
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(lineList, utteranceList)
# parse score
utterance_durations, bpm = scoreParser.csvDurationScoreParser(score_file)
# create the ground truth lab files
numOnset = 0
for idx, list in enumerate(nestedUtteranceLists):
if int(bpm[idx]):
print 'Counting onset number ... ' + recording_name + ' phrase ' + str(idx + 1)
ul = list[1]
numOnsetLine = len(ul)-1 # we don't count the first onset
numOnset += numOnsetLine
listOnset += [[recording_name, numOnset]]
return listOnset
def phraseBoundaryWriter(textgrid_file, outputFilename):
'''
Write phrase boundary from textgrid into outputFilename, example: .syll.lab
:param textgrid_file:
:param outputFilename:
:return:
'''
# read phrase list and utterance list
lineList = textgridParser.textGrid2WordList(textgrid_file,
whichTier='line')
utteranceList = textgridParser.textGrid2WordList(textgrid_file,
whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(
lineList, utteranceList)
# phrase start, end time
nonEmptyLineList = []
for list in nestedUtteranceLists:
nonEmptyLineList.append(list[0])
boundaryLabWriter(nonEmptyLineList, outputFilename)
def dump_training_data_textgrid_helper(wav_path,
textgrid_path,
recording_name,
tier_parent=None,
tier_child=None):
"""
load audio, textgrid
:param wav_path:
:param textgrid_path:
:param artist_name:
:param recording_name:
:param tier_parent: parent tier can be line
:param tier_childL child tier can be syllable
:return:
"""
ground_truth_textgrid_file = os.path.join(textgrid_path,
recording_name + '.TextGrid')
wav_file = os.path.join(wav_path, recording_name + '.wav')
line_list = textGrid2WordList(ground_truth_textgrid_file,
whichTier=tier_parent)
utterance_list = textGrid2WordList(ground_truth_textgrid_file,
whichTier=tier_child)
# parse lines of groundtruth
nested_utterance_lists, num_lines, num_utterances = wordListsParseByLines(
line_list, utterance_list)
# load audio
log_mel = getMFCCBands2DMadmom(wav_file, fs, hopsize_t, channel=1)
return nested_utterance_lists, log_mel
def getBoundaryNumber(textgrid_path, score_path):
"""
output a list to show the syllable number for each aria,
the syllable number is extracted from the textgrid
the textgrid needs to have a score
:param textgrid_path:
:param score_path:
:return:
"""
listOnset = []
list_file_path_name = []
for file_path_name in os.walk(textgrid_path):
list_file_path_name.append(file_path_name)
list_artist_level_path = list_file_path_name[0][1]
for artist_path in list_artist_level_path:
textgrid_artist_path = join(textgrid_path, artist_path)
recording_names = [f for f in os.listdir(textgrid_artist_path) if os.path.isfile(join(textgrid_artist_path, f))]
for rn in recording_names:
rn = rn.split('.')[0]
groundtruth_textgrid_file = join(textgrid_path, artist_path, rn+'.TextGrid')
# if artist_path=='danAll' or artist_path=='laosheng':
# score_file = join(score_path, rn+'.csv')
# else:
score_file = join(score_path, artist_path, rn + '.csv')
if not os.path.isfile(score_file):
continue
# print(groundtruth_textgrid_file)
lineList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='line')
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(lineList,
utteranceList)
# parse score
_, utterance_durations, bpm = scoreParser.csvDurationScoreParser(score_file)
# create the ground truth lab files
numOnset = 0
for idx, list in enumerate(nestedUtteranceLists):
try:
if float(bpm[idx]):
print('Counting onset number ... ' + rn + ' phrase ' + str(idx + 1))
ul = list[1]
numOnsetLine = len(ul) - 1 # we don't count the first onset
numOnset += numOnsetLine
except IndexError:
print(idx, 'not exist for recording', rn)
listOnset += [[artist_path, rn, numOnset]]
return listOnset
def textgrid_syllable_phoneme_parser(textgrid_file, tier1, tier2):
"""
Parse the textgrid file,
:param textgrid_file: filename
:param tier1: syllable tier
:param tier2: phoneme tier
:return: syllable and phoneme lists
"""
line_list = textGrid2WordList(textgrid_file, whichTier='line')
syllable_list = textGrid2WordList(textgrid_file, whichTier=tier1)
phoneme_list = textGrid2WordList(textgrid_file, whichTier=tier2)
# parse lines of groundtruth
nested_syllable_lists, _, _ = wordListsParseByLines(
line_list, syllable_list)
nested_phoneme_lists, _, _ = wordListsParseByLines(line_list, phoneme_list)
return nested_syllable_lists, nested_phoneme_lists
def dump_feature_onset_helper(wav_path, textgrid_path, artist_name,
recording_name):
groundtruth_textgrid_file = os.path.join(textgrid_path, artist_name,
recording_name + '.TextGrid')
wav_file = os.path.join(wav_path, artist_name, recording_name + '.wav')
lineList = textGrid2WordList(groundtruth_textgrid_file, whichTier='line')
utteranceList = textGrid2WordList(groundtruth_textgrid_file,
whichTier='dianSilence')
phonemeList = textGrid2WordList(groundtruth_textgrid_file,
whichTier='details')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = wordListsParseByLines(
lineList, utteranceList)
nestedPhonemeLists, _, _ = wordListsParseByLines(lineList, phonemeList)
# load audio
mfcc = get_log_mel_madmom(wav_file, fs, hopsize_t, channel=1)
return nestedUtteranceLists, nestedPhonemeLists, mfcc, phonemeList
def lineWordCount(textgrid_file):
"""
:param textgrid_file: annotation file
:return: numLines, numWords, numDians
"""
numLines, numWords, numDians = 0, 0, 0
wordDurationList, dianDurationList = [], []
entireLine = textgridParser.textGrid2WordList(textgrid_file, whichTier="line")
entireWordList = textgridParser.textGrid2WordList(textgrid_file, whichTier="pinyin")
entireDianList = textgridParser.textGrid2WordList(textgrid_file, whichTier="dian")
# parser word list for each line,
if len(entireWordList):
nestedWordLists, numLines, numWords = textgridParser.wordListsParseByLines(entireLine, entireWordList)
wordDurationList = wordDuration(nestedWordLists)
if len(entireDianList):
nestedWordLists, numLines, numDians = textgridParser.wordListsParseByLines(entireLine, entireDianList)
dianDurationList = wordDuration(nestedWordLists)
return numLines, numWords, numDians, wordDurationList, dianDurationList
def dump_feature_onset_helper(lab, wav_path, textgrid_path, score_path,
artist_name, recording_name, feature_type):
"""
load or parse audio, textgrid
:param lab:
:param wav_path:
:param textgrid_path:
:param score_path:
:param artist_name:
:param recording_name:
:param feature_type:
:return:
"""
if not lab:
ground_truth_textgrid_file = os.path.join(textgrid_path, artist_name,
recording_name + '.TextGrid')
wav_file = os.path.join(wav_path, artist_name, recording_name + '.wav')
line_list = textGrid2WordList(ground_truth_textgrid_file,
whichTier='line')
utterance_list = textGrid2WordList(ground_truth_textgrid_file,
whichTier='dianSilence')
# parse lines of groundtruth
nested_utterance_lists, num_lines, num_utterances = wordListsParseByLines(
line_list, utterance_list)
else:
ground_truth_textgrid_file = os.path.join(textgrid_path, artist_name,
recording_name + '.lab')
wav_file = os.path.join(wav_path, artist_name, recording_name + '.mp3')
nested_utterance_lists = [
lab2WordList(ground_truth_textgrid_file, label=True)
]
# parse score
score_file = os.path.join(score_path, artist_name, recording_name + '.csv')
_, utterance_durations, bpm = csvDurationScoreParser(score_file)
# load audio
if feature_type == 'madmom':
mfcc = getMFCCBands2DMadmom(wav_file, fs, hopsize_t, channel=1)
else:
print(feature_type + ' is not exist.')
raise
return nested_utterance_lists, utterance_durations, bpm, mfcc
def phraseBoundaryWriter(textgrid_file, outputFilename):
"""
Write phrase boundary from textgrid into outputFilename, example: .syll.lab
:param textgrid_file:
:param outputFilename:
:return:
"""
# read phrase list and utterance list
lineList = textgridParser.textGrid2WordList(textgrid_file, whichTier="line")
utteranceList = textgridParser.textGrid2WordList(textgrid_file, whichTier="utterance")
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(lineList, utteranceList)
# phrase start, end time
nonEmptyLineList = []
for list in nestedUtteranceLists:
nonEmptyLineList.append(list[0])
boundaryLabWriter(nonEmptyLineList, outputFilename)
def batch_eval(annotation_path,
segSyllable_path,
score_path,
groundtruth_path,
eval_details_path,
recordings,
tolerance,
label=False,
decoding_method='viterbi'):
sumDetectedBoundaries, sumGroundtruthPhrases, sumGroundtruthBoundaries, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion = 0 ,0 ,0 ,0 ,0 ,0, 0, 0
for artist_path, recording_name in recordings:
if annotation_path:
groundtruth_textgrid_file = os.path.join(
annotation_path, artist_path, recording_name + '.TextGrid')
groundtruth_lab_file_head = os.path.join(groundtruth_path,
artist_path)
else:
groundtruth_syllable_lab = os.path.join(groundtruth_path,
artist_path,
recording_name + '.lab')
detected_lab_file_head = os.path.join(segSyllable_path, artist_path,
recording_name)
score_file = os.path.join(score_path, artist_path,
recording_name + '.csv')
# parse score
if annotation_path:
_, _, utterance_durations, bpm = scoreParser.csvScorePinyinParser(
score_file)
else:
_, utterance_durations, bpm = scoreParser.csvDurationScoreParser(
score_file)
if eval_details_path:
eval_result_details_file_head = os.path.join(
eval_details_path, artist_path)
if not os.path.isfile(score_file):
print 'Score not found: ' + score_file
continue
if annotation_path:
# create ground truth lab path, if not exist
if not os.path.isdir(groundtruth_lab_file_head):
os.makedirs(groundtruth_lab_file_head)
if not os.path.isdir(eval_result_details_file_head):
os.makedirs(eval_result_details_file_head)
lineList = textgridParser.textGrid2WordList(
groundtruth_textgrid_file, whichTier='line')
utteranceList = textgridParser.textGrid2WordList(
groundtruth_textgrid_file, whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(
lineList, utteranceList)
# create the ground truth lab files
for idx, list in enumerate(nestedUtteranceLists):
try:
print(bpm[idx])
except IndexError:
continue
if float(bpm[idx]):
print 'Creating ground truth lab ... ' + recording_name + ' phrase ' + str(
idx + 1)
ul = list[1]
firstStartTime = ul[0][0]
groundtruthBoundaries = [
(np.array(ul_element[:2]) - firstStartTime).tolist() +
[ul_element[2]] for ul_element in ul
]
groundtruth_syllable_lab = join(
groundtruth_lab_file_head,
recording_name + '_' + str(idx + 1) + '.syll.lab')
with open(groundtruth_syllable_lab, "wb") as text_file:
for gtbs in groundtruthBoundaries:
text_file.write("{0} {1} {2}\n".format(
gtbs[0], gtbs[1], gtbs[2]))
else:
nestedUtteranceLists = [
labParser.lab2WordList(groundtruth_syllable_lab, label=label)
]
for idx, list in enumerate(nestedUtteranceLists):
try:
print(bpm[idx])
except IndexError:
continue
if float(bpm[idx]):
print 'Evaluating... ' + recording_name + ' phrase ' + str(
idx + 1)
if annotation_path:
ul = list[1]
firstStartTime = ul[0][0]
groundtruthBoundaries = [
(np.array(ul_element[:2]) - firstStartTime).tolist() +
[ul_element[2]] for ul_element in ul
]
else:
firstStartTime = list[0][0]
groundtruthBoundaries = [
(np.array(ul_element[:2]) - firstStartTime).tolist() +
[ul_element[2]] for ul_element in list
]
detected_syllable_lab = detected_lab_file_head + '_' + str(
idx + 1) + '.syll.lab'
if not os.path.isfile(detected_syllable_lab):
print 'Syll lab file not found: ' + detected_syllable_lab
continue
# read boundary detected lab into python list
lab_label = True if decoding_method == 'viterbi' else False
detectedBoundaries = labParser.lab2WordList(
detected_syllable_lab, label=lab_label)
numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, \
numInsertion, numDeletion, correct_list = evaluation2.boundaryEval(groundtruthBoundaries, detectedBoundaries, tolerance, label)
sumDetectedBoundaries += numDetectedBoundaries
sumGroundtruthBoundaries += numGroundtruthBoundaries
sumGroundtruthPhrases += 1
sumCorrect += numCorrect
sumOnsetCorrect += numOnsetCorrect
sumOffsetCorrect += numOffsetCorrect
sumInsertion += numInsertion
sumDeletion += numDeletion
# if numCorrect/float(numGroundtruthBoundaries) < 0.7:
print "Detected: {0}, Ground truth: {1}, Correct: {2}, Onset correct: {3}, " \
"Offset correct: {4}, Insertion: {5}, Deletion: {6}\n".\
format(numDetectedBoundaries, numGroundtruthBoundaries,numCorrect, numOnsetCorrect,
numOffsetCorrect, numInsertion, numDeletion)
return sumDetectedBoundaries, sumGroundtruthBoundaries, sumGroundtruthPhrases, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion
def batch_eval(aCapella_root,
dataset_path,
annotation_path,
segPhrase_path,
segSyllable_path,
score_path,
recordings,
tolerance,
label=True):
sumDetectedBoundaries, sumGroundtruthPhrases, sumGroundtruthBoundaries, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion = 0 ,0 ,0 ,0 ,0 ,0, 0, 0
for i_recording, recording_name in enumerate(recordings):
groundtruth_textgrid_file = os.path.join(aCapella_root, dataset_path,
annotation_path,
recording_name + '.TextGrid')
phrase_boundary_lab_file = os.path.join(aCapella_root, dataset_path,
segPhrase_path,
recording_name + '.lab')
# syll-o-matic output
# detected_lab_file_head = os.path.join(aCapella_root, dataset_path, segSyllable_path,recording_name)
# jan output
detected_lab_file_head = os.path.join(segSyllable_path, dataset_path,
recording_name)
score_file = os.path.join(aCapella_root, dataset_path, score_path,
recording_name + '.csv')
groundtruth_lab_file_head = os.path.join(aCapella_root, dataset_path,
groundtruth_lab_path,
recording_name)
eval_result_details_file_head = os.path.join(aCapella_root,
dataset_path,
eval_details_path,
recording_name)
if not os.path.isfile(score_file):
print 'Score not found: ' + score_file
continue
# create ground truth lab path, if not exist
if not os.path.isdir(groundtruth_lab_file_head):
os.makedirs(groundtruth_lab_file_head)
if not os.path.isdir(eval_result_details_file_head):
os.makedirs(eval_result_details_file_head)
lineList = textgridParser.textGrid2WordList(groundtruth_textgrid_file,
whichTier='line')
utteranceList = textgridParser.textGrid2WordList(
groundtruth_textgrid_file, whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(
lineList, utteranceList)
# parse score
utterance_durations, bpm = scoreParser.csvDurationScoreParser(
score_file)
# create the ground truth lab files
for idx, list in enumerate(nestedUtteranceLists):
if int(bpm[idx]):
print 'Creating ground truth lab ... ' + recording_name + ' phrase ' + str(
idx + 1)
ul = list[1]
firstStartTime = ul[0][0]
groundtruthBoundaries = [
(np.array(ul_element[:2]) - firstStartTime).tolist() +
[ul_element[2]] for ul_element in ul
]
groundtruth_syllable_lab = groundtruth_lab_file_head + '_' + str(
idx + 1) + '.syll.lab'
with open(groundtruth_syllable_lab, "wb") as text_file:
for gtbs in groundtruthBoundaries:
text_file.write("{0} {1} {2}\n".format(
gtbs[0], gtbs[1], gtbs[2]))
# syllable boundaries groundtruth of each line
# eval_details_csv = eval_result_details_file_head+'.csv'
# with open(eval_details_csv, 'wb') as csv_file:
# csv_writer = csv.writer(csv_file)
for idx, list in enumerate(nestedUtteranceLists):
if int(bpm[idx]):
print 'Evaluating... ' + recording_name + ' phrase ' + str(
idx + 1)
ul = list[1]
firstStartTime = ul[0][0]
groundtruthBoundaries = [
(np.array(ul_element[:2]) - firstStartTime).tolist() +
[ul_element[2]] for ul_element in ul
]
detected_syllable_lab = detected_lab_file_head + '_' + str(
idx + 1) + '.syll.lab'
if not os.path.isfile(detected_syllable_lab):
print 'Syll lab file not found: ' + detected_syllable_lab
continue
# read boundary detected lab into python list
detectedBoundaries = labParser.lab2WordList(
detected_syllable_lab, withLabel=label)
#
numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, \
numInsertion, numDeletion, correct_list = evaluation2.boundaryEval(groundtruthBoundaries, detectedBoundaries, tolerance, label)
sumDetectedBoundaries += numDetectedBoundaries
sumGroundtruthBoundaries += numGroundtruthBoundaries
sumGroundtruthPhrases += 1
sumCorrect += numCorrect
sumOnsetCorrect += numOnsetCorrect
sumOffsetCorrect += numOffsetCorrect
sumInsertion += numInsertion
sumDeletion += numDeletion
if numCorrect / float(numGroundtruthBoundaries) < 0.7:
print "Detected: {0}, Ground truth: {1}, Correct: {2}, Onset correct: {3}, " \
"Offset correct: {4}, Insertion: {5}, Deletion: {6}\n".\
format(numDetectedBoundaries, numGroundtruthBoundaries,numCorrect, numOnsetCorrect,
numOffsetCorrect, numInsertion, numDeletion)
# csv_writer.writerow([recording_name+'_'+str(idx+1),
# numDetectedBoundaries,
# numGroundtruthBoundaries,
# numCorrect,
# numInsertion,
# numDeletion,
# correct_list])
return sumDetectedBoundaries, sumGroundtruthBoundaries, sumGroundtruthPhrases, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion
def batch_eval(aCapella_root, dataset_path, annotation_path, segPhrase_path, segSyllable_path, score_path, recordings, tolerance):
sumDetectedBoundaries, sumGroundtruthPhrases, sumGroundtruthBoundaries, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion = 0 ,0 ,0 ,0 ,0 ,0, 0, 0
for i_recording, recording_name in enumerate(recordings):
groundtruth_textgrid_file = os.path.join(aCapella_root, dataset_path, annotation_path, recording_name+'.TextGrid')
phrase_boundary_lab_file = os.path.join(aCapella_root, dataset_path, segPhrase_path, recording_name+'.lab')
detected_lab_file_head = os.path.join(aCapella_root, dataset_path, segSyllable_path,recording_name)
score_file = os.path.join(aCapella_root, dataset_path, score_path, recording_name+'.csv')
if not os.path.isfile(score_file):
print 'Score not found: ' + score_file
continue
lineList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='line')
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='dianSilence')
# parse lines of groundtruth
nestedUtteranceLists, numLines, numUtterances = textgridParser.wordListsParseByLines(lineList, utteranceList)
# parse score
utterance_durations, bpm = scoreParser.csvDurationScoreParser(score_file)
# syllable boundaries groundtruth of each line
for idx, list in enumerate(nestedUtteranceLists):
if int(bpm[idx]):
print 'Evaluating... ' + recording_name + ' phrase ' + str(idx+1)
ul = list[1]
firstStartTime = ul[0][0]
groundtruthBoundaries = [(np.array(ul_element[:2]) - firstStartTime).tolist() for ul_element in ul]
detected_syllable_lab = detected_lab_file_head+'_'+str(idx+1)+'.syll.lab'
if not os.path.isfile(detected_syllable_lab):
print 'Syll lab file not found: ' + detected_syllable_lab
continue
# read boundary detected lab into python list
detectedBoundaries = labParser.lab2WordList(detected_syllable_lab)
# read boundary groundtruth textgrid into python list
# for segment in utteranceList:
# asciiLine = segment[2].encode("ascii", "replace")
# if len(asciiLine.replace(" ", "")):
# groundtruthBoundaries.append(segment[0:2])
#
# print groundtruthBoundaries
#
numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, \
numInsertion, numDeletion = evaluation.boundaryEval(groundtruthBoundaries, detectedBoundaries, tolerance)
sumDetectedBoundaries += numDetectedBoundaries
sumGroundtruthBoundaries += numGroundtruthBoundaries
sumGroundtruthPhrases += 1
sumCorrect += numCorrect
sumOnsetCorrect += numOnsetCorrect
sumOffsetCorrect += numOffsetCorrect
sumInsertion += numInsertion
sumDeletion += numDeletion
if numCorrect/float(numGroundtruthBoundaries) < 0.7:
print "Detected: {0}, Ground truth: {1}, Correct: {2}, Onset correct: {3}, " \
"Offset correct: {4}, Insertion: {5}, Deletion: {6}\n".\
format(numDetectedBoundaries, numGroundtruthBoundaries,numCorrect, numOnsetCorrect,
numOffsetCorrect, numInsertion, numDeletion)
return sumDetectedBoundaries, sumGroundtruthBoundaries, sumGroundtruthPhrases, sumCorrect, sumOnsetCorrect, \
sumOffsetCorrect, sumInsertion, sumDeletion
