def dimensional_scale_data(self):
self.ops.parseDescriptors()
dimscaling.buildarray_all_timevarying_descriptors(self.tgt.segs + self.cps.postLimitSegmentNormList, self.ops._normalizeDescriptors, self.AnalInterface)
from userclasses import SearchPassOptionsEntry as spass
from userclasses import SingleDescriptor as d
self.ops.SEARCH = [
spass('closest_percent', d('effDur-seg', norm=1), d('power-seg', norm=1), percent=25),
spass('closest', d('X', norm=1), d('Y', norm=1))
]
def descriptListPackageExpansion(initialListOfDescriptorObjs, numbMfccs):
from userclasses import SingleDescriptor as d
newListOfDescriptorObjs = []
for dobj in initialListOfDescriptorObjs:
metricsToWrite = []
if dobj.name.find('mfccs') != -1:
mfccWeightMask = np.power(np.linspace(1., 0.05, num=numbMfccs-1), 0.5)
for i in range(numbMfccs-1): metricsToWrite.append(('mfcc'+str(i+1)+dobj.name[5:], dobj.weight*(mfccWeightMask[i]/sum(mfccWeightMask)) ))
elif dobj.name.find('chromas') != -1:
for i in range(12): metricsToWrite.append( ( 'chroma'+str(i)+dobj.name[9:], dobj.weight/12. ) )
elif dobj.name.find('autocorrs') != -1:
for i in range(12): metricsToWrite.append( ( 'autocorr'+str(i)+dobj.name[9:], dobj.weight/12. ) )
elif dobj.name.find('crests') != -1:
for i in range(4):
metricsToWrite.append( [ 'crest'+str(i)+dobj.name[6:], dobj.weight/4. ] )
elif dobj.name.find('flatnesses') != -1:
for i in range(4):
metricsToWrite.append( [ 'flatness'+str(i)+dobj.name[10:], dobj.weight/4. ] )
elif dobj.name.find('harmonictristimuluses') != -1:
for i in range(3):
metricsToWrite.append( [ 'harmonictristimulus'+str(i)+dobj.name[21:], dobj.weight/3. ] )
elif dobj.name.find('perceptualtristimuluses') != -1:
for i in range(3):
metricsToWrite.append( [ 'perceptualtristimulus'+str(i)+dobj.name[23:], dobj.weight/3. ] )
# add original if not a package, add package elements otherwise
if len(metricsToWrite) == 0:
newListOfDescriptorObjs.append(dobj)
else:
for name, weight in metricsToWrite:
newdobj = d(name, weight=weight, norm=dobj.norm, normmethod=dobj.normmethod, distance=dobj.distance, limit=dobj.limit, energyWeight=dobj.energyWeight, origin=dobj.origin, neededBy=dobj.neededBy, packagename=dobj.name)
#print "ADDING", newdobj, newdobj.weight
newListOfDescriptorObjs.append(newdobj)
return newListOfDescriptorObjs
def addDescriptorIfNeeded(self, dobjToCheck, ops, addParents=False): from userclasses import SingleDescriptor as d if dobjToCheck.name not in [dn.name for dn in self.requiredDescriptors]: self.requiredDescriptors.append(dobjToCheck) if addParents: for pname in dobjToCheck.parents: self.addDescriptorIfNeeded(d(pname, origin=dobjToCheck.origin+'_PARENT'), ops)
def setupConcate(self, AnalInterface):
from userclasses import SingleDescriptor as d
self.powerStats = getDescriptorStatistics(self.segs, d('power'))
# initalise mixture and add norm coeffs to mixables...
for seg in self.segs:
seg.seek = 0
seg.selectiondone = False
seg.initMixture(AnalInterface)
for dobj in AnalInterface.mixtureDescriptors:
# copy the normalization paramaters from the target segments onto the mix descriptors
seg.mixdesc[dobj.name].setNorm(seg.desc[dobj.name].normdict)
def setupConcate(self, mixturelist):
from userclasses import SingleDescriptor as d
self.powerStats = getDescriptorStatistics(self.segs, d('power'))
# initalise mixture and add norm coeffs to mixables...
for seg in self.segs:
seg.seek = 0
seg.selectiondone = False
seg.classification = 0
seg.numberSelectedUnits = 0
seg.desc.rewind()
seg.desc.init_mixture(mixturelist)
seg.has_been_mixed = False
seg.originalPeak = seg.desc.get(
'peakTime-seg') # keep original, unsubtacted peak
def __init__(self, pm2_bin=None, supervp_bin=None, userWinLengthSec=0.12, userHopLengthSec=0.02, userEnergyHopLengthSec=0.005, resampleRate=12500, windowType='blackman', F0MaxAnalysisFreq=3000, F0MinFrequency=200, F0MaxFrequency=1000, F0AmpThreshold=30, numbMfccs=13, forceAnal=False, p=None, searchPaths=[], dataDirectoryLocation=None):
global descriptToFiles
# establish data directory
if dataDirectoryLocation == None:
self.dataDirectory = os.path.dirname(__file__)
else:
self.dataDirectory = os.path.abspath(dataDirectoryLocation)
if not os.path.exists(self.dataDirectory): os.makedirs(self.dataDirectory)
self.rawData = {} # file, descriptdict
self.p = p
self.ircamdescriptor_bin = os.path.join( os.path.dirname(__file__), 'ircamdescriptor-2.8.6', 'ircamdescriptor-2.8.6' )
assert os.path.exists(self.ircamdescriptor_bin)
# check for other bin files #
self.pm2_bin = findbin(pm2_bin, 'AudioSculpt3.0b7/Kernels/pm2')
self.supervp_bin = findbin(supervp_bin, 'AudioSculpt3.0b7/Kernels/supervp')
# anal
self.resampleRate = resampleRate
#############################################################################
## ensure that window and hop sizes are powers of two of the resample rate ##
#############################################################################
powersOfTwo = np.array([2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144, 524288])
closestWinSize = powersOfTwo[np.argmin(np.abs(powersOfTwo-(userWinLengthSec*float(self.resampleRate))))]
closestHopSize = powersOfTwo[np.argmin(np.abs(powersOfTwo-(userHopLengthSec*float(self.resampleRate))))]
#self.winLengthSec = closestWinSize/float(self.resampleRate) # adjusted value to esure its a power of two in the resample rate!
#self.hopLengthSec = closestHopSize/float(self.resampleRate) # adjusted value to esure its a power of two in the resample rate!
self.winLengthSec = userWinLengthSec
self.hopLengthSec = userHopLengthSec
self.userEnergyWinLengthSec = userEnergyHopLengthSec*2
self.userEnergyHopLengthSec = userEnergyHopLengthSec
self.IOBufferSize = min(4096, closestWinSize*4, self.userEnergyWinLengthSec*float(self.resampleRate)*4)
self.windowType = windowType
self.F0MaxAnalysisFreq = F0MaxAnalysisFreq
self.F0MinFrequency = F0MinFrequency
self.F0MaxFrequency = F0MaxFrequency
self.F0AmpThreshold = F0AmpThreshold
# set up mfccs
self.numbMfccs = numbMfccs
for i in range(self.numbMfccs):
descriptToFiles.append(("mfcc%i"%i, 'ircamd', False, True, 'MFCC', self.numbMfccs, i))
# other stuff
self.forceAnal = forceAnal
self.searchPaths = searchPaths
### setup ircam binary config text file
self.analdir = os.path.join(self.dataDirectory, 'data')
self.ircamd_infofile = os.path.join(self.analdir, 'ircamd-info.json')
self.ircamd_info = {}
if not os.path.exists(self.analdir): os.makedirs(self.analdir)
### setup anal registry
self.dataRegistryPath = os.path.join(self.analdir, 'data_registry.json')
if os.path.exists(self.dataRegistryPath):
fh = open(self.dataRegistryPath)
self.dataRegistry = json.load(fh)
fh.close()
else:
self.dataRegistry = {}
self.internalDescriptorNames = ['power', 'peakTime-seg', 'power-seg', 'power-mean-seg', 'effDurFrames-seg', 'effDur-seg','f0-seg', 'MIDIPitch-seg']
self.config_loc = os.path.join(self.analdir, 'ircamdescriptor.config.txt')
self.config_text = '''; DO NOT EDIT - AUTOMATICALLY GENERATED BY anallinkage.py
[Parameters]
ResampleTo = %i
NormalizeSignal = 0
SubtractMean = 1
IOBufferSize = %i
SaveShortTermTMFeatures = 1
WindowType = %s
OutputFormat = raw
F0MaxAnalysisFreq = %i
F0MinFrequency = %i
F0MaxFrequency = %i
F0AmpThreshold = %i
MFCCs = %i
[StandardDescriptors]
WindowSize = %f
HopSize = %f
TextureWindowsFrames = -1
TextureWindowsHopFrames = -1
Harmonics = 20
DeviationStopBand = 10
RolloffThreshold = 0.95
DecreaseThreshold = 0.4
NoiseThreshold = 0.15
;~~~~~~~~~~~~~~~~~~~descriptors~~~~~~~~~~~~~~~~~~~~~
SignalZeroCrossingRate = ShortTime
;AutoCorrelation = ShortTime
TotalEnergy = ShortTime
SpectralCentroid = ShortTime
SpectralSpread = ShortTime
SpectralSkewness = ShortTime
SpectralKurtosis = ShortTime
SpectralSlope = ShortTime
SpectralDecrease = ShortTime
SpectralRolloff = ShortTime
SpectralVariation = ShortTime
Loudness = ShortTime
Spread = ShortTime
Sharpness = ShortTime
PerceptualSpectralCentroid = ShortTime
PerceptualSpectralSpread = ShortTime
PerceptualSpectralSkewness = ShortTime
PerceptualSpectralKurtosis = ShortTime
PerceptualSpectralSlope = ShortTime
PerceptualSpectralDecrease = ShortTime
PerceptualSpectralRolloff = ShortTime
PerceptualSpectralVariation = ShortTime
PerceptualSpectralDeviation = ShortTime
PerceptualOddToEvenRatio = ShortTime
PerceptualTristimulus = ShortTime
MFCC = ShortTime
SpectralFlatness = ShortTime
SpectralCrest = ShortTime
FundamentalFrequency = ShortTime
NoiseEnergy = ShortTime
Noisiness = ShortTime
Inharmonicity = ShortTime
HarmonicEnergy = ShortTime
HarmonicSpectralCentroid = ShortTime
HarmonicSpectralSpread = ShortTime
HarmonicSpectralSkewness = ShortTime
HarmonicSpectralKurtosis = ShortTime
HarmonicSpectralSlope = ShortTime
HarmonicSpectralDecrease = ShortTime
HarmonicSpectralRolloff = ShortTime
HarmonicSpectralVariation = ShortTime
HarmonicSpectralDeviation = ShortTime
HarmonicOddToEvenRatio = ShortTime
HarmonicTristimulus = ShortTime
Chroma = ShortTime
[EnergyDescriptors]
WindowSize = %f
HopSize = %f
TextureWindowsFrames = -1
TextureWindowsHopFrames = -1
;~~~~~~~~~~~~~~~~energy descriptors~~~~~~~~~~~~~~~~
EnergyEnvelope = 1
'''%(self.resampleRate, self.IOBufferSize, self.windowType, self.F0MaxAnalysisFreq, self.F0MinFrequency, self.F0MaxFrequency, self.F0AmpThreshold, self.numbMfccs, self.winLengthSec, self.hopLengthSec, self.userEnergyWinLengthSec, self.userEnergyHopLengthSec)
# make a list of all possible descriptor objects
self.allDescriptors = []
from userclasses import SingleDescriptor as d
for desc in descriptToFiles:
dobj = d(desc[0])
self.allDescriptors.append( dobj )
if not dobj.seg: self.allDescriptors.append( d(desc[0]+'-seg') )
for desc in self.internalDescriptorNames:
self.allDescriptors.append( d(desc) )
# print all descriptors
#print (', '.join(["d('%s')"%d.name for d in self.allDescriptors if d.seg]))
# write log
if self.p != None:
self.p.log("ANALYSIS CONFIG: using analysis window of %.3f (%i samples)"%(self.winLengthSec, closestWinSize))
self.p.log("ANALYSIS CONFIG: using analysis overlap of %.3f (%i samples)"%(self.hopLengthSec, closestHopSize))
def getDescriptorLists(self, ops): self.expandDescriptorPackages(ops) from userclasses import SingleDescriptor as d from audioguide.userclasses import SearchPassOptionsEntry as spassObj self.requiredDescriptors = [] # add SEARCH descriptors for spass in ops.SEARCH: for dobj in spass.descriptor_list: #dobj.origin = 'SEARCH' self.addDescriptorIfNeeded(dobj, ops, addParents=True) # add target onset descriptors for dname, weight in self.tgtOnsetDescriptors.items(): self.addDescriptorIfNeeded(d(dname, weight=weight, origin='TARGET_ONSET'), ops) # add limiting descriptors if 'limit' in ops.CORPUS_GLOBAL_ATTRIBUTES: for stringy in ops.CORPUS_GLOBAL_ATTRIBUTES['limit']: print(stringy.split()[0], 'GLOBAL_LIMIT') self.addDescriptorIfNeeded(d(stringy.split()[0], origin='GLOBAL_LIMIT'), ops, addParents=True) if hasattr(ops, 'CORPUS'): for csfObj in ops.CORPUS: for stringy in csfObj.limit: self.addDescriptorIfNeeded(d(stringy.split()[0], origin='LOCAL_LIMIT'), ops, addParents=True) # add EXPERIMENTAL spass entries for k, v in ops.EXPERIMENTAL.items(): if isinstance(v, spassObj): for dobj in v.descriptor_list: dobj.origin = 'EXPERIMENTAL' self.addDescriptorIfNeeded(dobj, ops, addParents=True) # add CLUSTER descriptors if 'descriptors' in ops.CLUSTER_MAPPING: for s in ops.CLUSTER_MAPPING['descriptors']: self.addDescriptorIfNeeded(d(s+'-seg', origin='CLUSTER_MAPPING'), ops, addParents=True) # add classification descriptors if 'descriptors' in ops.OUTPUTEVENT_CLASSIFY: for s in ops.OUTPUTEVENT_CLASSIFY['descriptors']: self.addDescriptorIfNeeded(d(s, origin='CLASSIFICATION'), ops, addParents=True) # add segmentation data descriptor if ops.SEGMENTATION_FILE_INFO != 'logic': self.addDescriptorIfNeeded(d(ops.SEGMENTATION_FILE_INFO, weight=0, origin='SEGMENTATION_DATA'), ops, addParents=True) # add ordering by descriptor if None not in [ops.ORDER_CORPUS_BY_DESCRIPTOR]: self.addDescriptorIfNeeded(d(ops.ORDER_CORPUS_BY_DESCRIPTOR, weight=0, origin='ORDER_CORPUS_BY_DESCRIPTOR'), ops, addParents=True) for dname, weight in self.tgtOnsetDescriptors.items(): self.addDescriptorIfNeeded(d(dname, weight=weight, origin='TARGET_ONSET'), ops) # add internal mectrics if not already used for dname in self.internalDescriptorNames: self.addDescriptorIfNeeded(d(dname, origin='INTERNAL'), ops, addParents=True) # # # make normalisation list! self.normalizeDescriptors = [] for dobj in self.requiredDescriptors: if dobj.origin in ['SEARCH', 'EXPERIMENTAL']: self.normalizeDescriptors.append(dobj) # make mixture list! self.mixtureDescriptors = [] tmpmix = ['power'] for dobj in self.requiredDescriptors: if dobj.origin == 'SEARCH' and dobj.is_mixable: if dobj.seg: tmpmix.append(dobj.name) # make sure segs are at the end else: tmpmix.insert(0, dobj.name) for pobjname in dobj.parents: tmpmix.insert(0, pobjname) for dname in tmpmix: for dobj in self.requiredDescriptors: if dobj.name == dname: self.mixtureDescriptors.append(dobj) break