def pitchtrackGetter(fname, outputPitchtrack):
# pitchContours object
print 'extracting pitch contours ... ...'
pitchMakam = pc.PitchExtractMakam()
pitchMakam.setup()
contours_bins, contours_contourSaliences, contours_start_times, duration = pitchMakam.run(fname)
contours_start_times = contours_start_times.tolist()
# filter short pitch contours
print 'filtering short pitch contours ... ...'
contours_bins, contours_contourSaliences, contours_start_times, toRmIndexCF = cr.contourFilter(contours_bins, contours_contourSaliences, contours_start_times, framesize = 2048, hopsize = 128)
# feature extraction
print 'extracting features ... ...'
lengthContour, meanPitchContour, sdPitchContour, totalSalience, meanSalience, sdSalience, mfccs = fc.featureExtract(contours_bins, contours_contourSaliences)
# load mean sd for feature normalization
meansdpath = './model/meansdJson.json'
with open(meansdpath) as data_file:
data = json.load(data_file)
meansd = data['meansd']
# normalize the feature vector
featureVec = []
for ii in range(len(lengthContour)):
f1 = [(lengthContour[ii]/meansd[0][0])/meansd[0][1], (meanPitchContour[ii]/meansd[1][0])/meansd[1][1], (sdPitchContour[ii]/meansd[2][0])/meansd[2][1], (totalSalience[ii]/meansd[3][0])/meansd[3][1], (meanSalience[ii]/meansd[4][0])/meansd[4][1], (sdSalience[ii]/meansd[5][0])/meansd[5][1]]
for kk in range(len(mfccs[ii])):
f2.append((mfccs[ii][kk]-meansd[6+kk][0])/meansd[6+kk][1])
feature = f1 + f2
featureVec.append(feature)
# classifying pitch contours
print 'classifying pitch contours ... ...'
knn = joblib.load('./model/jingju_pitchContoursClassificationModel.pkl')
classification = knn.predict(featureVec) # 0: voice 1: nonvoice
classification = classification.tolist()
contours_bins_out = []
contours_contourSaliences_out = []
contours_start_times_out = []
for ii in range(len(classification)):
if classification[ii] == 0:
contours_bins_out.append(contours_bins[ii])
contours_contourSaliences_out.append(contours_contourSaliences[ii])
contours_start_times_out.append(contours_start_times[ii])
# dumping contours into pitch track
out = pitchMakam.pitchtrackDumper(contours_bins_out, contours_contourSaliences_out, contours_start_times_out, duration)
np.savetxt(outputPitchtrack, out)
return
def voiceContoursGetter(pitchContourJson, classificationResult):
contours_bins, contours_contourSaliences, contours_start_times, duration = cr.contourReader(pitchContourJson)
contours_bins, contours_contourSaliences, contours_start_times, toRmIndex_cal = cr.contourFilter(contours_bins, contours_contourSaliences, contours_start_times, framesize = 2048, hopsize = 128)
toRmIndex_json, classification = classificationResultReader(classificationResult)
if len(toRmIndex_cal) == len(toRmIndex_json):
contours_bins_out = []
contours_contourSaliences_out = []
contours_start_times_out = []
for ii in range(len(classification)):
if classification[ii] == 0:
contours_bins_out.append(contours_bins[ii])
contours_contourSaliences_out.append(contours_contourSaliences[ii])
contours_start_times_out.append(contours_start_times[ii])
return contours_bins_out, contours_contourSaliences_out, contours_start_times_out, duration
else:
print 'problem with toRmIndex length, not equal.'
return
def featureNormalization(featureList):
'''
the structure of normalized feature list is [(feature0, mean0, sd0), (feature1, mean1, sd1), ...]
'''
normalizedFeatureList = []
for ii in range(len(featureList)-1):
normalizedFeature, mean, sd = fc.scaleFeatures(featureList[ii])
normalizedFeatureList.append((normalizedFeature, mean, sd))
normalizedFeatureList.append(featureList[-1])
return normalizedFeatureList
jj = 1
for filename in filenamesvoicePath:
print jj, len(allfilenames)
contours_bins, contours_contourSaliences, contours_start_times, duration = cr.contourReader(filename)
contours_bins, contours_contourSaliences, contours_start_times, toRmIndexCF = cr.contourFilter(contours_bins, contours_contourSaliences, contours_start_times, framesize = 2048, hopsize = 128)
#contours_bins, contours_contourSaliences, contours_start_times, toRmIndexRF = registerFilter(contours_bins, contours_contourSaliences, contours_start_times, tonicHz = 350.0) # need to give tonic firstly
lengthContour, meanPitchContour, sdPitchContour, totalSalience, meanSalience, sdSalience, mfccs = fc.featureExtract(contours_bins, contours_contourSaliences)
featureList = featureListCreate(featureList, lengthContour, meanPitchContour, sdPitchContour, totalSalience, meanSalience, sdSalience, mfccs, 0)
jj += 1
for filename in filenamesnonvoicePath:
print jj, len(allfilenames)
contours_bins, contours_contourSaliences, contours_start_times, duration = cr.contourReader(filename)
contours_bins, contours_contourSaliences, contours_start_times, toRmIndexCF = cr.contourFilter(contours_bins, contours_contourSaliences, contours_start_times, framesize = 2048, hopsize = 128)
#contours_bins, contours_contourSaliences, contours_start_times, toRmIndexRF = registerFilter(contours_bins, contours_contourSaliences, contours_start_times, tonicHz = 350.0) # need to give tonic firstly
lengthContour, meanPitchContour, sdPitchContour, totalSalience, meanSalience, sdSalience, mfccs = fc.featureExtract(contours_bins, contours_contourSaliences)