def resynthesizePitch(praatEXE,
inputWavFN,
pitchFN,
outputWavFN,
minPitch,
maxPitch,
scriptFN=None,
pointList=None):
'''
Resynthesizes the pitch in a wav file with the given pitch contour file
The pitch track to use can optionally be passed in as pointList. If
so, it will be saved as pitchFN for praat to be able to use.
'''
if scriptFN is None:
scriptFN = join(utils.scriptsPath, "resynthesize_pitch.praat")
if pointList is not None:
dur = audioio.WavQueryObj(inputWavFN).getDuration()
pointObj = dataio.PointObject2D(pointList, dataio.PITCH, 0, dur)
pointObj.save(pitchFN)
utils.runPraatScript(
praatEXE, scriptFN,
[inputWavFN, pitchFN, outputWavFN, minPitch, maxPitch])
def changeDuration(fromWavFN, durationParameters, stepList, outputName,
outputMinPitch, outputMaxPitch, praatEXE):
'''
Uses praat to morph duration in one file to duration in another
Praat uses the PSOLA algorithm
'''
rootPath = os.path.split(fromWavFN)[0]
# Prep output directories
outputPath = join(rootPath, "duration_resynthesized_wavs")
utils.makeDir(outputPath)
durationTierPath = join(rootPath, "duration_tiers")
utils.makeDir(durationTierPath)
fromWavDuration = audio_scripts.getSoundFileDuration(fromWavFN)
durationParameters = copy.deepcopy(durationParameters)
# Pad any gaps with values of 1 (no change in duration)
# No need to stretch out any pauses at the beginning
if durationParameters[0][0] != 0:
tmpVar = (0, durationParameters[0][0] - PRAAT_TIME_DIFF, 1)
durationParameters.insert(0, tmpVar)
# Or the end
if durationParameters[-1][1] < fromWavDuration:
durationParameters.append(
(durationParameters[-1][1] + PRAAT_TIME_DIFF, fromWavDuration, 1))
# Create the praat script for doing duration manipulation
for stepAmount in stepList:
durationPointList = []
for start, end, ratio in durationParameters:
percentChange = 1 + (ratio - 1) * stepAmount
durationPointList.append((start, percentChange))
durationPointList.append((end, percentChange))
outputPrefix = "%s_%0.3g" % (outputName, stepAmount)
durationTierFN = join(durationTierPath,
"%s.DurationTier" % outputPrefix)
outputWavFN = join(outputPath, "%s.wav" % outputPrefix)
durationTier = dataio.PointObject2D(durationPointList, dataio.DURATION,
0, fromWavDuration)
durationTier.save(durationTierFN)
praat_scripts.resynthesizeDuration(praatEXE, fromWavFN, durationTierFN,
outputWavFN, outputMinPitch,
outputMaxPitch)
outputWavFN = "mary1_accented.wav"
outputPitchFN = "mary1_accented.pitch"
minPitch = 75
maxPitch = 450
if not os.path.exists(rootOutputPath):
os.mkdir(rootOutputPath)
# 1st - get pitch
piList = pitch_and_intensity.extractPI(join(root, wavFN),
join(rootOutputPath, pitchIntensityFN),
praatEXE, minPitch, maxPitch)
pitchList = [(timeV, pitchV) for timeV, pitchV, _ in piList]
dur = audioio.WavQueryObj(join(root, wavFN)).getDuration()
pointObj = dataio.PointObject2D(pitchList, dataio.PITCH, 0, dur)
pointObj.save(join(rootOutputPath, originalPitchFN))
# 2nd - get region to manipulate. Let's make the subject more emphatic!
tg = tgio.openTextgrid(join(root, "mary1.TextGrid"))
tier = tg.tierDict["words"]
start, stop, _ = tier.entryList[0] # Getting info for the first word
targetPitchList = [(timeV, pitchV) for timeV, pitchV in pitchList
if timeV >= start and timeV <= stop]
# 3rd - make manipulation
accent = modify_pitch_accent.PitchAccent(targetPitchList)
accent.addPlateau(0.05) # Peak is dragged out for 0.05 seconds
accent.adjustPeakHeight(60) # Plateau is raised by 60 hz
accent.shiftAccent(-0.03) # Recenter the plateau around the original peak
def f0Morph(fromWavFN,
pitchPath,
stepList,
outputName,
doPlotPitchSteps,
fromPitchData,
toPitchData,
outputMinPitch,
outputMaxPitch,
praatEXE,
keepPitchRange=False,
keepAveragePitch=False):
'''
Resynthesizes the pitch track from a source to a target wav file
fromPitchData and toPitchData should be segmented according to the
portions that you want to morph. The two lists must have the same
number of sublists.
Occurs over a three-step process.
This function can act as a template for how to use the function
morph_sequence.morphChunkedDataLists to morph pitch contours or
other data.
By default, everything is morphed, but it is possible to maintain elements
of the original speaker's pitch (average pitch and pitch range) by setting
the appropriate flag)
'''
fromDuration = audio_scripts.getSoundFileDuration(fromWavFN)
# Iterative pitch tier data path
pitchTierPath = join(pitchPath, "pitchTiers")
resynthesizedPath = join(pitchPath, "f0_resynthesized_wavs")
for tmpPath in [pitchTierPath, resynthesizedPath]:
utils.makeDir(tmpPath)
# 1. Prepare the data for morphing - acquire the segments to merge
# (Done elsewhere, with the input fed into this function)
# 2. Morph the fromData to the toData
try:
finalOutputList = morph_sequence.morphChunkedDataLists(
fromPitchData, toPitchData, stepList)
except IndexError:
raise MissingPitchDataException()
fromPitchData = [row for subList in fromPitchData for row in subList]
toPitchData = [row for subList in toPitchData for row in subList]
# 3. Save the pitch data and resynthesize the pitch
mergedDataList = []
for i in range(0, len(finalOutputList)):
outputDataList = finalOutputList[i]
if keepPitchRange is True:
outputDataList = morph_sequence.morphRange(outputDataList,
fromPitchData)
if keepAveragePitch is True:
outputDataList = morph_sequence.morphAveragePitch(
outputDataList, fromPitchData)
stepOutputName = "%s_%0.3g" % (outputName, stepList[i])
pitchFNFullPath = join(pitchTierPath, "%s.PitchTier" % stepOutputName)
outputFN = join(resynthesizedPath, "%s.wav" % stepOutputName)
pointObj = dataio.PointObject2D(outputDataList, dataio.PITCH, 0,
fromDuration)
pointObj.save(pitchFNFullPath)
outputTime, outputVals = list(zip(*outputDataList))
mergedDataList.append((outputTime, outputVals))
praat_scripts.resynthesizePitch(praatEXE, fromWavFN, pitchFNFullPath,
outputFN, outputMinPitch,
outputMaxPitch)
# 4. (Optional) Plot the generated contours
if doPlotPitchSteps:
fromTime, fromVals = list(zip(*fromPitchData))
toTime, toVals = list(zip(*toPitchData))
plot_morphed_data.plotF0((fromTime, fromVals), (toTime, toVals),
mergedDataList,
join(pitchTierPath, "%s.png" % outputName))
def f0Morph(fromWavFN,
pitchPath,
stepList,
outputName,
doPlotPitchSteps,
fromPitchData,
toPitchData,
outputMinPitch,
outputMaxPitch,
praatEXE,
keepPitchRange=False,
keepAveragePitch=False,
sourcePitchDataList=None,
minIntervalLength=0.3):
'''
Resynthesizes the pitch track from a source to a target wav file
fromPitchData and toPitchData should be segmented according to the
portions that you want to morph. The two lists must have the same
number of sublists.
Occurs over a three-step process.
This function can act as a template for how to use the function
morph_sequence.morphChunkedDataLists to morph pitch contours or
other data.
By default, everything is morphed, but it is possible to maintain elements
of the original speaker's pitch (average pitch and pitch range) by setting
the appropriate flag)
sourcePitchDataList: if passed in, any regions unspecified by
fromPitchData will be sampled from this list. In
essence, this allows one to leave segments of
the original pitch contour untouched by the
morph process.
'''
fromDuration = audio_scripts.getSoundFileDuration(fromWavFN)
# Find source pitch samples that will be mixed in with the target
# pitch samples later
nonMorphPitchData = []
if sourcePitchDataList is not None:
timeList = sorted(fromPitchData)
timeList = [(row[0][0], row[-1][0]) for row in timeList]
endTime = sourcePitchDataList[-1][0]
invertedTimeList = praatio_utils.invertIntervalList(timeList, endTime)
invertedTimeList = [(start, stop) for start, stop in invertedTimeList
if stop - start > minIntervalLength]
for start, stop in invertedTimeList:
pitchList = praatio_utils.getValuesInInterval(
sourcePitchDataList, start, stop)
nonMorphPitchData.extend(pitchList)
# Iterative pitch tier data path
pitchTierPath = join(pitchPath, "pitchTiers")
resynthesizedPath = join(pitchPath, "f0_resynthesized_wavs")
for tmpPath in [pitchTierPath, resynthesizedPath]:
utils.makeDir(tmpPath)
# 1. Prepare the data for morphing - acquire the segments to merge
# (Done elsewhere, with the input fed into this function)
# 2. Morph the fromData to the toData
try:
finalOutputList = morph_sequence.morphChunkedDataLists(
fromPitchData, toPitchData, stepList)
except IndexError:
raise MissingPitchDataException()
fromPitchData = [row for subList in fromPitchData for row in subList]
toPitchData = [row for subList in toPitchData for row in subList]
# 3. Save the pitch data and resynthesize the pitch
mergedDataList = []
for i in range(0, len(finalOutputList)):
outputDataList = finalOutputList[i]
if keepPitchRange is True:
outputDataList = morph_sequence.morphRange(outputDataList,
fromPitchData)
if keepAveragePitch is True:
outputDataList = morph_sequence.morphAveragePitch(
outputDataList, fromPitchData)
if sourcePitchDataList is not None:
outputDataList.extend(nonMorphPitchData)
outputDataList.sort()
stepOutputName = "%s_%0.3g" % (outputName, stepList[i])
pitchFNFullPath = join(pitchTierPath, "%s.PitchTier" % stepOutputName)
outputFN = join(resynthesizedPath, "%s.wav" % stepOutputName)
pointObj = dataio.PointObject2D(outputDataList, dataio.PITCH, 0,
fromDuration)
pointObj.save(pitchFNFullPath)
outputTime, outputVals = zip(*outputDataList)
mergedDataList.append((outputTime, outputVals))
praat_scripts.resynthesizePitch(praatEXE, fromWavFN, pitchFNFullPath,
outputFN, outputMinPitch,
outputMaxPitch)
# 4. (Optional) Plot the generated contours
if doPlotPitchSteps:
fromTime, fromVals = zip(*fromPitchData)
toTime, toVals = zip(*toPitchData)
plot_morphed_data.plotF0((fromTime, fromVals), (toTime, toVals),
mergedDataList,
join(pitchTierPath, "%s.png" % outputName))