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 g_eval(groundtruth_textgrid_filename, georgi_alignment_filename, tolerance):
boundaryList = georgiParser.syllables_total_parser(georgi_alignment_filename)
try:
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dian', utf16=False)
utteranceDuration = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dianDuration', utf16=False)
except:
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dian', utf16=True)
utteranceDuration = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dianDuration', utf16=True)
# remove empty dian
tempGroundtruthList = []
groundtruthDuration = []
for idx, utterance in enumerate(utteranceList):
if len(utterance[2].strip()):# and float(utteranceDuration[idx][2]):
tempGroundtruthList.append(utterance)
groundtruthDuration.append(float(utteranceDuration[idx][2]))
# remove 0 duration
detectedBoundaryList = []
groundtruthList = []
for idx, utterance in enumerate(tempGroundtruthList):
if groundtruthDuration[idx]:
groundtruthList.append(tempGroundtruthList[idx])
detectedBoundaryList.append(boundaryList[idx])
numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, numInsertion, numDeletion, correct_list = \
evaluation2.boundaryEval(groundtruthList,detectedBoundaryList,tolerance)
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 numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, numInsertion, numDeletion
def g_eval(groundtruth_textgrid_filename, georgi_alignment_filename, tolerance):
boundaryList = georgiParser.syllables_total_parser(georgi_alignment_filename)
try:
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dian', utf16=False)
utteranceDuration = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dianDuration', utf16=False)
except:
utteranceList = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dian', utf16=True)
utteranceDuration = textgridParser.textGrid2WordList(groundtruth_textgrid_filename, whichTier='dianDuration', utf16=True)
# remove empty dian
tempGroundtruthList = []
groundtruthDuration = []
for idx, utterance in enumerate(utteranceList):
if len(utterance[2].strip()):# and float(utteranceDuration[idx][2]):
tempGroundtruthList.append(utterance)
groundtruthDuration.append(float(utteranceDuration[idx][2]))
# remove 0 duration
detectedBoundaryList = []
groundtruthList = []
for idx, utterance in enumerate(tempGroundtruthList):
if groundtruthDuration[idx]:
groundtruthList.append(tempGroundtruthList[idx])
detectedBoundaryList.append(boundaryList[idx])
numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, numInsertion, numDeletion = \
evaluation.boundaryEval(groundtruthList,detectedBoundaryList,tolerance)
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 numDetectedBoundaries, numGroundtruthBoundaries, numCorrect, numOnsetCorrect, numOffsetCorrect, numInsertion, numDeletion
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 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 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 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 lyrics_textgrid_csvwriter(filenames_textgrid,
textgrid_path,
path_to_save,
extension=''):
for fn in filenames_textgrid:
fn_full = os.path.join(textgrid_path, fn + extension)
list_line = textGrid2WordList(fn_full, whichTier='line')
fn_csv = os.path.join(path_to_save, fn + '.csv')
with open(fn_csv, 'w', newline='') as csvfile:
writer = csv.writer(csvfile, delimiter=';')
for line in list_line:
if len(line[2]):
writer.writerow([line[2]])
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
''' This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Author: Rong Gong, MTG-UPF, [email protected] 17 January 2016 ''' import textgridParser groundtruth_textgrid_file = '../laosheng/lsxp-Wo_zheng_zai-Kong_cheng_ji01-upf.TextGrid' # parse the phrase boundary, and its content lineList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='line') # parse the dian Tier dianList = textgridParser.textGrid2WordList(groundtruth_textgrid_file, whichTier='dian') print lineList print dianSilenceList
