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 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
