def test_AlignmentIo_readLines(self, its=50):
for it in range(its):
numLevels = randint(1, 4)
alignment = gen_alignment(numLevels=numLevels)
framePeriod = gen_framePeriod()
alignmentIo = alio.AlignmentIo(framePeriod)
alignmentLines = alignmentIo.writeLines(alignment)
alignmentAgain = alignmentIo.readLines(alignmentLines)
self.assertEqual(alignmentAgain, alignment)
def test_AlignmentIo_writeLines_1_level(self, its=50):
for it in range(its):
framePeriod = gen_framePeriod()
alignment = gen_alignment()
simpleAlignmentIo = alio.SimpleAlignmentIo(framePeriod)
alignmentIo = alio.AlignmentIo(framePeriod)
self.assertEqual(
simpleAlignmentIo.writeLines(alignment),
alignmentIo.writeLines(alignment)
)
def test_AlignmentIo_writeLines_nested(self, its=50):
for it in range(its):
numLevels = randint(2, 4)
alignment = gen_alignment(numLevels=numLevels)
framePeriod = gen_framePeriod()
alignmentIo = alio.AlignmentIo(framePeriod)
alignmentLines = alignmentIo.writeLines(alignment)
alignmentLinesAgain = []
for startTime, endTime, label, subAlignment in alignment:
subAlignmentLines = alignmentIo.writeLines(subAlignment)
alignmentLinesAgain.append(subAlignmentLines[0] + ' ' + label)
alignmentLinesAgain.extend(subAlignmentLines[1:])
self.assertEqual(alignmentLinesAgain, alignmentLines)
def main(argv):
parser = argparse.ArgumentParser(description=__doc__.split('\n')[0])
parser.add_argument('--alignment_suffix',
dest='alignmentSuffix',
metavar='ALIGNSUFFIX',
default='lab',
help='suffix for alignment files')
parser.add_argument(dest='labelMapFile',
metavar='LABELMAP',
help=('file specifying label mapping'
' (two columns separated by whitespace)'))
parser.add_argument(dest='alignmentDirIn',
metavar='ALIGNDIRIN',
help='directory to read input alignments from')
parser.add_argument(
dest='uttIdsFile',
metavar='UTTIDSFILE',
help=('file containing a list of utterance ids'
' (e.g. one line might be "cmu_us_arctic_slt_a0001")'))
parser.add_argument(dest='alignmentDirOut',
metavar='ALIGNDIROUT',
help='directory to write output alignments to')
args = parser.parse_args(argv[1:])
uttIds = [line.strip() for line in open(args.uttIdsFile)]
alignmentIo = alio.AlignmentIo(framePeriod=1e-7)
labelMapDict = readLabelMapFile(args.labelMapFile)
print '(read label map with %s entries)' % len(labelMapDict)
print '(writing output to directory %s)' % args.alignmentDirOut
for uttId in uttIds:
alignmentFileIn = os.path.join(args.alignmentDirIn,
'%s.%s' % (uttId, args.alignmentSuffix))
alignmentFileOut = os.path.join(
args.alignmentDirOut, '%s.%s' % (uttId, args.alignmentSuffix))
alignment = alignmentIo.readFile(alignmentFileIn)
alignmentNew = mapAlignment(labelMapDict, alignment)
alignmentIo.writeFile(alignmentFileOut, alignmentNew)
def main(rawArgs):
parser = argparse.ArgumentParser(
description=(
'Computes the MCD metric for two sequences of mel cepstra.'
' Mel cepstral distortion (MCD) is a measure of the difference'
' between two sequences of mel cepstra.'
' This utility computes the MCD between two sequences of equal'
' length that are assumed to already be "aligned" in terms of'
' their timing.'
' Optionally certain segments (e.g. silence) can be omitted from'
' the calculation.'),
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument(
'--ext',
dest='ext',
default='mgc',
metavar='EXT',
help=('file extension added to uttId to get file containing speech'
' parameters'))
parser.add_argument('--param_order',
dest='paramOrder',
default=40,
type=int,
metavar='ORDER',
help='parameter order of the cepstral files')
parser.add_argument('--remove_segments',
dest='removeSegments',
default=None,
metavar='LABELREGEX',
help='regex of segment labels to remove')
parser.add_argument(
'--alignment_dir',
dest='alignmentDir',
default=None,
metavar='ALIGNMENTDIR',
help=('directory containing phone-level alignment files (used for'
' segment removal)'))
parser.add_argument(
'--frame_period',
dest='framePeriod',
default=0.005,
type=float,
metavar='FRAMEPERIOD',
help='frame period in seconds (used for segment removal)')
parser.add_argument(dest='natDir',
metavar='NATDIR',
help='directory containing natural speech parameters')
parser.add_argument(
dest='synthDir',
metavar='SYNTHDIR',
help='directory containing synthetic speech parameters')
parser.add_argument(dest='uttIds',
metavar='UTTID',
nargs='+',
help='utterance ids (ext will be appended to these)')
args = parser.parse_args(rawArgs[1:])
assert (args.removeSegments is None) == (args.alignmentDir is None)
if args.removeSegments is not None:
print(('NOTE: removing segments matching regex \'%s\' using alignments'
' in %s' % (args.removeSegments, args.alignmentDir)))
reRemoveSegments = (None if args.removeSegments is None else re.compile(
args.removeSegments))
costFn = mt.logSpecDbDist
alignmentIo = alio.AlignmentIo(args.framePeriod)
getAlignment = DirReader(alignmentIo, args.alignmentDir, 'lab')
vecSeqIo = vsio.VecSeqIo(args.paramOrder)
getNatVecSeq = DirReader(vecSeqIo, args.natDir, args.ext)
getSynthVecSeq = DirReader(vecSeqIo, args.synthDir, args.ext)
costTot = 0.0
framesTot = 0
for uttId in args.uttIds:
print(('processing', uttId))
nat = getNatVecSeq(uttId)
synth = getSynthVecSeq(uttId)
# ignore 0th cepstral component
nat = nat[:, 1:]
synth = synth[:, 1:]
assert len(nat) == len(synth)
if reRemoveSegments is None:
cost = sum([
costFn(natFrame, synthFrame)
for natFrame, synthFrame in zip(nat, synth)
])
frames = len(nat)
else:
alignment = getAlignment(uttId)
alignmentInclude = [(startTime, endTime,
not reRemoveSegments.search(label))
for startTime, endTime, label, _ in alignment]
includeFrames = list(util.expandAlignment(alignmentInclude))
assert len(includeFrames) == len(nat)
costs = [
costFn(natFrame, synthFrame)
for natFrame, synthFrame, includeFrame in zip(
nat, synth, includeFrames) if includeFrame
]
cost = sum(costs)
frames = len(costs)
costTot += cost
framesTot += frames
print(('overall MCD = %f (%d frames)' % (costTot / framesTot, framesTot)))
def main(argv):
parser = argparse.ArgumentParser(description=__doc__.split('\n')[0])
parser.add_argument('--alignment_suffix',
dest='alignmentSuffix',
metavar='ALIGNSUFFIX',
default='lab',
help='suffix for alignment files (e.g. "lab")')
parser.add_argument(
'--sublabel_pat',
dest='subLabelStrEndPat',
metavar='PAT',
default='[%d]',
help=('printf-style pattern which, when expanded by replacing %%d with'
' a sublabel index >= 2, specifies the last part of the sublabel'
' string used in the input alignment files'
' (e.g. "[%%d]")'))
parser.add_argument(
'--num_sublabels',
dest='numSubLabels',
metavar='NUMSUBLABELS',
default=5,
type=int,
help=('number of sublabels (also sometimes known as "states") in input'
' alignments (e.g. "5")'))
parser.add_argument(
dest='quesFile',
metavar='QUESFILE',
help='HTK / HTS question file (e.g. "questions_qst001.hed")')
parser.add_argument(dest='alignmentDirIn',
metavar='ALIGNDIRIN',
help='directory to read input alignments from')
parser.add_argument(
dest='uttIdsFile',
metavar='UTTIDSFILE',
help=('file containing a list of utterance ids'
' (e.g. one line might be "cmu_us_arctic_slt_a0001")'))
parser.add_argument(dest='alignmentDirOut',
metavar='ALIGNDIROUT',
help='directory to write output alignments to')
args = parser.parse_args(argv[1:])
uttIds = [line.strip() for line in open(args.uttIdsFile)]
subLabelStrEnds = [
args.subLabelStrEndPat % (subLabelIndex + 2)
for subLabelIndex in range(args.numSubLabels)
]
alignmentIo = alio.AlignmentIo(framePeriod=1e-7)
questions = qio.readQuesFileVerifying(args.quesFile)
quesRes = [qio.getQuesRe(quesPats) for _, quesPats in questions]
print '(writing output to directory %s)' % args.alignmentDirOut
for uttId in uttIds:
alignmentFileIn = os.path.join(args.alignmentDirIn,
'%s.%s' % (uttId, args.alignmentSuffix))
alignmentFileOut = os.path.join(
args.alignmentDirOut, '%s.%s' % (uttId, args.alignmentSuffix))
alignment = alignmentIo.readFile(alignmentFileIn)
alignmentNew = mapAlignment(alignment, quesRes, subLabelStrEnds)
alignmentIo.writeFile(alignmentFileOut, alignmentNew)
def main(argv):
parser = argparse.ArgumentParser(description=__doc__.split('\n')[0])
parser.add_argument('--alignment_suffix',
dest='alignmentSuffix',
metavar='ALIGNSUFFIX',
default='lab',
help='suffix for alignment files (e.g. "lab")')
parser.add_argument(
'--sublabel_pat',
dest='subLabelStrEndPat',
metavar='PAT',
default='[%d]',
help=('printf-style pattern which, when expanded by replacing %%d with'
' a sublabel index >= 2, specifies the last part of the sublabel'
' string used in the input alignment files'
' (e.g. "[%%d]")'))
parser.add_argument(
'--stream_pat_pat',
dest='streamSpecPat',
metavar='PATPAT',
default='{*}[%d].stream[1]',
help=('printf-style pattern which, when expanded by replacing %%d with'
' a sublabel index >= 2, specifies the HTK macro pattern'
' identifying the desired tree for a given sublabel'
' (e.g. for HTS demo a value of "{*}[%%d].stream[1]" would use'
' trees from stream 1, which is the mgc stream)'))
parser.add_argument(
'--num_sublabels',
dest='numSubLabels',
metavar='NUMSUBLABELS',
default=5,
type=int,
help=('number of sublabels (also sometimes known as "states") in input'
' alignments (e.g. "5")'))
parser.add_argument(
dest='treeFile',
metavar='TREE',
help='HTS demo-style decision tree file (e.g. "mgc.inf")')
parser.add_argument(dest='alignmentDirIn',
metavar='ALIGNDIRIN',
help='directory to read input alignments from')
parser.add_argument(
dest='uttIdsFile',
metavar='UTTIDSFILE',
help=('file containing a list of utterance ids'
' (e.g. one line might be "cmu_us_arctic_slt_a0001")'))
parser.add_argument(dest='alignmentDirOut',
metavar='ALIGNDIROUT',
help='directory to write output alignments to')
args = parser.parse_args(argv[1:])
uttIds = [line.strip() for line in open(args.uttIdsFile)]
subLabelStrEnds = [
args.subLabelStrEndPat % (subLabelIndex + 2)
for subLabelIndex in range(args.numSubLabels)
]
alignmentIo = alio.AlignmentIo(framePeriod=1e-7)
def subLabelIndexToStreamSpec(subLabelIndex):
return args.streamSpecPat % (subLabelIndex + 2)
questions, streamSpecedTrees = tio.readTreeFileVerifying(args.treeFile)
quesReDict = qio.getQuesReDict(questions)
navTrees = []
for subLabelIndex in range(args.numSubLabels):
streamSpecDesired = subLabelIndexToStreamSpec(subLabelIndex)
foundTrees = [
tree for streamSpec, tree in streamSpecedTrees
if streamSpec == streamSpecDesired
]
assert len(foundTrees) == 1
navTree = tio.NavBinaryTree(quesReDict, foundTrees[0])
navTrees.append(navTree)
numLeaves = sum([len(navTree.tree.leaves) for navTree in navTrees])
print '(found %s leaves)' % numLeaves
print '(writing output to directory %s)' % args.alignmentDirOut
for uttId in uttIds:
alignmentFileIn = os.path.join(args.alignmentDirIn,
'%s.%s' % (uttId, args.alignmentSuffix))
alignmentFileOut = os.path.join(
args.alignmentDirOut, '%s.%s' % (uttId, args.alignmentSuffix))
alignment = alignmentIo.readFile(alignmentFileIn)
alignmentNew = mapAlignment(alignment, navTrees, subLabelStrEnds)
alignmentIo.writeFile(alignmentFileOut, alignmentNew)