def __init__(self):
# classes = [0,1,2,3,4,5] # detail classes
nc1 = nc.noteClass()
self.classes = nc1.segmentClasses.values()
self.featureNamesDict = nc1.featureNamesDict
self.resetFeatureDictAll()
def __init__(self):
self.nc1 = nc.noteClass()
self.ptSeg1 = ptSeg.pitchtrackSegByNotes()
self.evalu1 = evalu.Evaluation()
self.evaluationMetrics = []
self.resetRepresentation()
def constructJson4DetailFeature(self, featureVecFolderPath,
targetFolderPath, recordingNames,
groundtruthCSVPath):
# detail level classification groundtruth json
nc1 = nc.noteClass()
segmentClasses = nc1.segmentClasses
for rn in recordingNames:
featureFilename = featureVecFolderPath + rn + '.json'
targetFilename = targetFolderPath + rn + '.json'
CSVFile = ''.join([str(groundtruthCSVPath), '/', rn, '.csv'])
with open(featureFilename) as data_file:
featureDict = json.load(data_file)
# read classification groundtruth csv file
with open(CSVFile) as groundtruthFile:
targetDict = {}
for line in groundtruthFile.readlines():
splitLine = line.split(',')
for iic in range(len(splitLine)):
c = splitLine[iic].rstrip('\n')
if c != '': # empty string is False
cInt = int(c)
targetDict[cInt] = iic
with open(targetFilename, 'w') as outfile:
json.dump(targetDict, outfile)
def __init__(self):
self.nc1 = nc.noteClass()
self.ptSeg1 = ptSeg.pitchtrackSegByNotes()
self.evalu1 = evalu.Evaluation()
self.evaluationMetrics = []
self.resetRepresentation()
def __init__(self):
# classes = [0,1,2,3,4,5] # detail classes
nc1 = nc.noteClass()
self.classes = nc1.segmentClasses.values()
self.featureNamesDict = nc1.featureNamesDict
self.resetFeatureDictAll()
def constructJson4NoteFeature(self,featureVecFolderPath,targetFolderPath,recordingNames,groundtruthNoteLevel):
# note level classification groundtruth json
nc1 = nc.noteClass()
allNoteClasses = list(nc1.basicNoteClasses.keys()) + nc1.unClassifiedNoteClasses
for rn in recordingNames:
featureFilename = featureVecFolderPath+rn+'.json'
targetFilename = targetFolderPath+rn+'.json'
with open(featureFilename) as data_file:
featureDict = json.load(data_file)
targetDict = {}
for ii in range(1, len(featureDict)+1):
for noteClass in allNoteClasses:
noteClassRecordingFoldername = os.path.join(groundtruthNoteLevel, noteClass, rn+'midinote')
if os.path.isdir(noteClassRecordingFoldername):
onlypngs = [ f for f in os.listdir(noteClassRecordingFoldername) if f.endswith('.png') ]
for png in onlypngs:
pngNum = os.path.splitext(png)[0]
if str(ii) == pngNum:
if noteClass in nc1.basicNoteClasses:
# targetDict[ii] = nc1.basicNoteClasses[noteClass] # detail classes
targetDict[ii] = 0 # 5 basic classes
else:
# targetDict[ii] = 5 # detail classes
targetDict[ii] = 1 # non classified classes
with open(targetFilename, 'w') as outfile:
json.dump(targetDict, outfile)
def constructJson4DetailFeature(self, featureVecFolderPath,targetFolderPath,recordingNames,groundtruthCSVPath):
# detail level classification groundtruth json
nc1 = nc.noteClass()
segmentClasses = nc1.segmentClasses
for rn in recordingNames:
featureFilename = featureVecFolderPath+rn+'.json'
targetFilename = targetFolderPath+rn+'.json'
CSVFile = ''.join([str(groundtruthCSVPath),'/',rn,'.csv'])
with open(featureFilename) as data_file:
featureDict = json.load(data_file)
# read classification groundtruth csv file
with open(CSVFile) as groundtruthFile:
targetDict = {}
for line in groundtruthFile.readlines():
splitLine = line.split(',')
for iic in range(len(splitLine)):
c = splitLine[iic].rstrip('\n')
if c != '': # empty string is False
cInt = int(c)
targetDict[cInt] = iic
with open(targetFilename, 'w') as outfile:
json.dump(targetDict, outfile)
def constructJson4NoteFeature(self, featureVecFolderPath, targetFolderPath,
recordingNames, groundtruthNoteLevel):
# note level classification groundtruth json
nc1 = nc.noteClass()
allNoteClasses = list(
nc1.basicNoteClasses.keys()) + nc1.unClassifiedNoteClasses
for rn in recordingNames:
featureFilename = featureVecFolderPath + rn + '.json'
targetFilename = targetFolderPath + rn + '.json'
with open(featureFilename) as data_file:
featureDict = json.load(data_file)
targetDict = {}
for ii in range(1, len(featureDict) + 1):
for noteClass in allNoteClasses:
noteClassRecordingFoldername = os.path.join(
groundtruthNoteLevel, noteClass, rn + 'midinote')
if os.path.isdir(noteClassRecordingFoldername):
onlypngs = [
f for f in os.listdir(noteClassRecordingFoldername)
if f.endswith('.png')
]
for png in onlypngs:
pngNum = os.path.splitext(png)[0]
if str(ii) == pngNum:
if noteClass in nc1.basicNoteClasses:
# targetDict[ii] = nc1.basicNoteClasses[noteClass] # detail classes
targetDict[ii] = 0 # 5 basic classes
else:
# targetDict[ii] = 5 # detail classes
targetDict[
ii] = 1 # non classified classes
with open(targetFilename, 'w') as outfile:
json.dump(targetDict, outfile)
def __init__(self):
self.nc1 = nc.noteClass()
def jingjuElementSegmentation():
start_time = time.time() # starting time
'''
######################################### pYIN pitchtrack and notes ################################################
filename_amateur = '/Users/gong/Documents/MTG document/Jingju arias/jingjuElementMaterials/laosheng/test/weiguojia_section_amateur.wav'
filename_pro = '/Users/gong/Documents/MTG document/Jingju arias/jingjuElementMaterials/laosheng/test/weiguojia_section_pro.wav'
pYINPtNote(filename_amateur)
#filename1 = pYinPath + '/testAudioLong.wav'
'''
############################################## initialsation #######################################################
pitchtrackNoteTrainFolderPath = os.path.join(dir, 'pYinOut/laosheng/train/')
# classification training ground truth
# groundtruthNoteLevelPath = '/Users/gong/Documents/pycharmProjects/jingjuSegPic/laosheng/train/pyinNoteCurvefit/classified'
# groundtruthNoteDetailPath = '/Users/gong/Documents/pycharmProjects/jingjuSegPic/laosheng/train/refinedSegmentCurvefit/classified'
# classification model path
pitchContourClassificationModelName = os.path.join(dir,'pYinOut/laosheng/train/model/pitchContourClassificationModel.pkl')
# feature, target folders train
# featureVecTrainFolderPath = os.path.join(dir,'pYinOut/laosheng/train/featureVec/')
# targetTrainFolderPath = os.path.join(dir,'pYinOut/laosheng/train/target/')
# predict folders
pitchtrackNotePredictFolderPath = os.path.join(dir,'../../data/updateFiles/')
#pitchtrackNotePredictFolderPath = os.path.join(dir,'pYinOut/laosheng/predict/)
featureVecPredictFolderPath = os.path.join(dir,'pYinOut/laosheng/predict/featureVec/')
targetPredictFolderPath = os.path.join(dir,'pYinOut/laosheng/predict/target/')
# recordingNamesTrain = ['male_02_neg_1', 'male_12_neg_1', 'male_12_pos_1', 'male_13_pos_1', 'male_13_pos_3']
recordingNamesPredict = ['student']
evaluation = False # parameters grid search
slopeTh = 60.0 # contour combination slope difference threshold
flatnoteTh = 80.0 # threshold for deciding one note as flat pitch note
#recordingNamesPredict = ['male_02_neg_1', 'male_12_neg_1', 'male_12_pos_1', 'male_13_pos_1', 'male_13_pos_3'] # evaluation
################################################## predict #########################################################
# segmentation
nc2 = nc.noteClass()
nc2.noteSegmentationFeatureExtraction(pitchtrackNotePredictFolderPath,
featureVecPredictFolderPath,
pitchtrackNotePredictFolderPath,
recordingNamesPredict,
segCoef=0.3137,predict=True)
# predict
ttknn2 = ttknn.TrainTestKNN()
ttknn2.predict(pitchContourClassificationModelName,featureVecPredictFolderPath,
targetPredictFolderPath,recordingNamesPredict)
########################################### representation #########################################################
for rm in recordingNamesPredict:
# filename declaration
originalPitchtrackFilename = pitchtrackNotePredictFolderPath+rm+'_pitchtrack.csv'
targetFilename = targetPredictFolderPath+rm+'.json'
refinedSegmentFeaturesFilename = pitchtrackNotePredictFolderPath+rm+'_refinedSegmentFeatures.json'
representationFilename = pitchtrackNotePredictFolderPath+rm+'_representation.json'
figureFilename = pitchtrackNotePredictFolderPath+rm+'_reprensentationContourFigure.png'
# important txt files!!
regressionPitchtrackFilename = pitchtrackNotePredictFolderPath+rm+'_regression_pitchtrack.csv'
refinedSegmentationGroundtruthFilename = pitchtrackNotePredictFolderPath+rm+'_refinedSeg.csv'
rsm1 = rsm.RefinedSegmentsManipulation()
rsm1.process(refinedSegmentFeaturesFilename,targetFilename,
representationFilename,figureFilename,regressionPitchtrackFilename,
originalPitchtrackFilename = originalPitchtrackFilename,
refinedSegGroundtruthFilename=refinedSegmentationGroundtruthFilename,
slopeTh=slopeTh, flatnoteTh=flatnoteTh)
runningTime = time.time() - start_time
# print("--- %s seconds ---" % (time.time() - start_time))
return runningTime
######### construct target json ###########
fvt1 = fvt.FeatureVecTarget()
fvt1.constructJson4DetailFeature(featureVecTrainFolderPath,targetTrainFolderPath,recordingNamesTrain,groundtruthNoteDetailPath)
################ train ####################
ttknn1 = ttknn.TrainTestKNN()
ttknn1.gatherFeatureTarget(featureVecTrainFolderPath,targetTrainFolderPath,recordingNamesTrain)
# ttknn1.featureVec2DPlot([3,7])
ttknn1.crossValidation(pitchContourClassificationModelName)
'''
################################################## predict #########################################################
# segmentation
nc2 = nc.noteClass()
nc2.noteSegmentationFeatureExtraction(pitchtrackNotePredictFolderPath,
featureVecPredictFolderPath,
recordingNamesPredict,
segCoef=0.3137,predict=True)
# predict
ttknn2 = ttknn.TrainTestKNN()
ttknn2.predict(pitchContourClassificationModelName,featureVecPredictFolderPath,
targetPredictFolderPath,recordingNamesPredict)
'''
################################################ evaluation code ###################################################
# uncomment it only when needs evaluation
with open('./pYinOut/laosheng/predict/evaluationResult02.txt', "w") as outfile:
for sc in np.linspace(0.2,0.5,30):
######### construct target json ###########
fvt1 = fvt.FeatureVecTarget()
fvt1.constructJson4DetailFeature(featureVecTrainFolderPath,targetTrainFolderPath,recordingNamesTrain,groundtruthNoteDetailPath)
################ train ####################
ttknn1 = ttknn.TrainTestKNN()
ttknn1.gatherFeatureTarget(featureVecTrainFolderPath,targetTrainFolderPath,recordingNamesTrain)
# ttknn1.featureVec2DPlot([3,7])
ttknn1.crossValidation(pitchContourClassificationModelName)
'''
################################################## predict #########################################################
# segmentation
nc2 = nc.noteClass()
nc2.noteSegmentationFeatureExtraction(pitchtrackNotePredictFolderPath,
featureVecPredictFolderPath,
recordingNamesPredict,
segCoef=0.3137,predict=True)
# predict
ttknn2 = ttknn.TrainTestKNN()
ttknn2.predict(pitchContourClassificationModelName,featureVecPredictFolderPath,
targetPredictFolderPath,recordingNamesPredict)
'''
################################################ evaluation code ###################################################
# uncomment it only when needs evaluation
with open('./pYinOut/laosheng/predict/evaluationResult02.txt', "w") as outfile:
for sc in np.linspace(0.2,0.5,30):
def __init__(self):
self.nc1 = nc.noteClass()
