def fill_evident_holes(self):
""" Fill holes if we find consecutive index values in prev/next anchors. """
to_add = []
for i in range(1, len(self)):
prevann = self[i - 1]
curann = self[i]
if prevann.GetLabel().IsSilence():
continue
if curann.GetLabel().IsSilence():
continue
# there is a hole
if prevann.GetLocation().GetEnd() < curann.GetLocation().GetBegin(
):
idxprev = prevann.GetLabel().GetTypedValue()
idxcur = curann.GetLabel().GetTypedValue()
prevend = prevann.GetLocation().GetEnd()
curbegin = curann.GetLocation().GetBegin()
if idxprev + 1 == idxcur - 1:
text = Text(idxprev + 1, data_type="int")
hole = Annotation(TimeInterval(prevend, curbegin),
Label(text))
to_add.append(hole)
for a in to_add:
self.Add(a)
return len(to_add)
def test_window(self):
self.t = AnchorTier()
self.t.set_duration(12.)
self.t.set_win_delay(4.)
f, t = self.t.fix_window(0.)
self.assertEqual(f, 0.)
self.assertEqual(t, 4.)
self.t.set_duration(18.)
self.t.Append(
Annotation(TimeInterval(TimePoint(13.), TimePoint(13.5)),
Label(Text(18, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.), TimePoint(14.6)),
Label(Text(20, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.6), TimePoint(15.6)),
Label(Text(21, data_type="int"))))
f, t = self.t.fix_window(14.)
self.assertEqual(f, 15.6)
self.assertEqual(t, 18.)
def test_export(self):
self.t = AnchorTier()
self.t.set_duration(17.8)
self.t.set_win_delay(4.)
self.t.set_ext_delay(1.)
self.t.set_out_delay(0.2)
# self.t.Append(Annotation(TimeInterval(TimePoint(0.), TimePoint(1.)), Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(1.5), TimePoint(2.)),
Label(Text(3, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(4.5), TimePoint(6.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(7.), TimePoint(8.)),
Label(Text(8, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(13.), TimePoint(13.5)),
Label(Text(12, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.), TimePoint(14.6)),
Label(Text(13, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.6), TimePoint(15.2)),
Label(Text(14, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(15.2), TimePoint(15.6)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(16.), TimePoint(16.60)),
Label(Text(16, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(16.63), TimePoint(17.)),
Label(Text(17, data_type="int"))))
toklist = [
u"w0", u"w1", u"w2", u"w3", u"w4", u"w5", u"w6", u"w7", u"w8",
u"w9", u"w10", u"w11", u"w12", u"w13", u"w14", u"w15", u"w16",
u"w17"
]
newtier = self.t.export(toklist)
def append_silences(self, channel):
""" Append silences as anchors.
:param channel: (sppasChannel)
"""
logging.debug(" ... Search silences:")
# We have to find tracks first
tracks_times = autils.search_channel_speech(channel, self._win_length,
self._min_sil_dur,
self._min_track_dur,
self._shift_dur_start,
self._shift_dur_end)
radius = self._win_length / 2.
toprec = 0.
# Then, the silences are the holes between tracks
for (from_time, to_time) in tracks_times:
if toprec < from_time:
begin = TimePoint(toprec, radius)
if begin == 0.:
begin = TimePoint(toprec, radius)
end = TimePoint(from_time, radius)
a = Annotation(TimeInterval(begin, end), Label("#"))
self.Append(a)
toprec = to_time
# A silence at the end?
if toprec < self._duration:
begin = TimePoint(toprec, radius)
end = TimePoint(self._duration, 0.)
a = Annotation(TimeInterval(begin, end), Label("#"))
self.Append(a)
for i, a in enumerate(self):
logging.debug(" ... ... %i: %s" % (i, a))
def test_holes(self):
self.t = AnchorTier()
self.t.set_duration(17.8)
self.t.set_win_delay(4.)
self.t.set_ext_delay(1.)
self.t.set_out_delay(0.2)
self.t.Append(
Annotation(TimeInterval(TimePoint(0.), TimePoint(1.5)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(1.5), TimePoint(2.)),
Label(Text(0, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(3.), TimePoint(3.5)),
Label(Text(-1, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(4.5), TimePoint(6.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(7.), TimePoint(8.)),
Label(Text(8, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(11.3), TimePoint(12.)),
Label(Text(-1, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(13.), TimePoint(13.5)),
Label(Text(18, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.), TimePoint(14.6)),
Label(Text(20, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.6), TimePoint(15.6)),
Label(Text(21, data_type="int"))))
self.assertTrue(self.t.check_holes_ntokens(10))
self.assertFalse(self.t.check_holes_ntokens(9))
self.assertEqual(self.t.fill_evident_holes(), 1)
def test_window_sil(self):
self.t = AnchorTier()
self.t.set_duration(17.8)
self.t.set_win_delay(4.)
self.t.set_ext_delay(1.)
self.t.set_out_delay(0.2)
self.t.Append(
Annotation(TimeInterval(TimePoint(0.), TimePoint(1.5)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(4.5), TimePoint(6.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(9.7), TimePoint(11.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.6), TimePoint(17.8)),
Label("#")))
f, t = self.t.fix_window(0.)
self.assertEqual(f, 1.5)
self.assertEqual(t, 4.5)
f, t = self.t.fix_window(0.5)
self.assertEqual(f, 1.5)
self.assertEqual(t, 4.5)
f, t = self.t.fix_window(1.5)
self.assertEqual(f, 1.5)
self.assertEqual(t, 4.5)
f, t = self.t.fix_window(3.5)
self.assertEqual(f, 3.5)
self.assertEqual(t, 4.5)
f, t = self.t.fix_window(4.5)
self.assertEqual(f, 6.3)
self.assertEqual(t, 9.7)
f, t = self.t.fix_window(9.)
self.assertEqual(f, 9.)
self.assertEqual(t, 9.7)
f, t = self.t.fix_window(9.6)
self.assertEqual(f, 11.3)
self.assertEqual(t, 14.6)
f, t = self.t.fix_window(14.6)
self.assertEqual(f, 17.8)
self.assertEqual(t, 17.8)
self.t.set_win_delay(10.)
f, t = self.t.fix_window(0)
self.assertEqual(f, 1.5)
self.assertEqual(t, 4.5)
self.t.set_win_delay(2.)
f, t = self.t.fix_window(0)
self.assertEqual(f, 1.5)
self.assertEqual(t, 3.5)
self.t.set_win_delay(1.)
f, t = self.t.fix_window(0)
self.assertEqual(f, 1.5)
self.assertEqual(t, 2.5)
def test_near(self):
self.t = AnchorTier()
self.t.set_duration(17.8)
self.t.set_win_delay(4.)
self.t.set_ext_delay(1.)
self.t.set_out_delay(0.2)
self.assertIsNone(self.t.near_indexed_anchor(1., -1))
self.t.Append(
Annotation(TimeInterval(TimePoint(0.), TimePoint(1.5)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(1.5), TimePoint(2.)),
Label(Text(1, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(3.), TimePoint(3.5)),
Label(Text(-1, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(4.5), TimePoint(6.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(7.), TimePoint(8.)),
Label(Text(2, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(9.7), TimePoint(11.3)),
Label("#")))
self.t.Append(
Annotation(TimeInterval(TimePoint(11.3), TimePoint(12.)),
Label(Text(-1, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.), TimePoint(14.6)),
Label(Text(3, data_type="int"))))
self.t.Append(
Annotation(TimeInterval(TimePoint(14.6), TimePoint(17.8)),
Label("#")))
a = self.t.Near(13., 1)
self.assertEqual(a, 7)
a = self.t.Near(17., 1)
self.assertEqual(a, -1)
a = self.t.Near(17.8, 1)
self.assertEqual(a, -1)
a = self.t.near_indexed_anchor(1., 1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(1.5, 1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(2., 1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(11., 1)
self.assertEqual(a.GetLabel().GetTypedValue(), 3)
a = self.t.near_indexed_anchor(15., 1)
self.assertIsNone(a)
a = self.t.near_indexed_anchor(1., -1)
self.assertIsNone(a)
a = self.t.near_indexed_anchor(3., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(3.5, -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(5., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(7., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 1)
a = self.t.near_indexed_anchor(8., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(9.7, -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(11., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(12., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(14., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 2)
a = self.t.near_indexed_anchor(18., -1)
self.assertEqual(a.GetLabel().GetTypedValue(), 3)
def export(self, toklist):
""" Create the "Chunks" tier and return it.
:param toklist: Tokens used to fill the intervals.
"""
tier = Tier("Chunks")
# Append silences in the result and pop them from the list of anchors
anchors = AnchorTier()
for ann in self:
if ann.GetLabel().IsSilence():
try:
tier.Append(ann.Copy())
except Exception:
logging.debug("Error: Silence not appended: %s" % ann)
pass
else:
try:
anchors.Append(ann.Copy())
except Exception:
logging.debug("Error: Anchor not appended: %s" % ann)
pass
# Fill the holes when prev-index and next-index made a sequence
anchors.fill_evident_holes()
# Fill holes between anchors
for i in range(1, len(anchors)):
prevann = anchors[i - 1]
curann = anchors[i]
# there is a hole
if prevann.GetLocation().GetEnd() < curann.GetLocation().GetBegin(
):
idxprev = prevann.GetLabel().GetTypedValue()
idxcur = curann.GetLabel().GetTypedValue()
prevend = prevann.GetLocation().GetEnd()
curbegin = curann.GetLocation().GetBegin()
if idxprev + 1 == idxcur:
# hum... a little bit of hack!!!
holeduration = curbegin.GetMidpoint(
) - prevend.GetMidpoint()
if holeduration < 0.055:
prevend.SetMidpoint(curbegin.GetMidpoint())
elif holeduration < 0.505 and prevann.GetLocation(
).GetDuration().GetValue() < 0.105:
prevend.SetMidpoint(curbegin.GetMidpoint())
elif (idxprev + 1) < idxcur - 1:
texte = " ".join(toklist[idxprev + 1:idxcur])
begin = prevend.GetMidpoint()
end = curbegin.GetMidpoint()
hole = Annotation(
TimeInterval(TimePoint(begin), TimePoint(end)),
Label(texte))
anns = tier.Find(begin, end, overlaps=True)
if len(anns) == 0:
try:
tier.Add(hole)
except Exception:
pass
# Append chunk of anchors
start = 0
end = 0
to_continue = True
if end + 1 >= anchors.GetSize():
to_continue = False
while to_continue:
idxcur = anchors[end].GetLabel().GetTypedValue()
idxnex = anchors[end + 1].GetLabel().GetTypedValue()
endtimecur = anchors[end].GetLocation().GetEnd()
begtimenex = anchors[end + 1].GetLocation().GetBegin()
# a sequence of anchors is finished if either:
# - next anchor index does not directly follow the current one
# - next anchor time does not directly follow the current one
# - we already appended enough anchors in the chunk
if idxcur + 1 != idxnex or (end -
start) > 10 or begtimenex > endtimecur:
# append the chunk
idxstart = anchors[start].GetLabel().GetTypedValue()
chunk_text = " ".join(toklist[idxstart:idxcur + 1])
tbegin = anchors[start].GetLocation().GetBegin()
tend = anchors[end].GetLocation().GetEnd()
ann = Annotation(TimeInterval(tbegin, tend), Label(chunk_text))
tier.Remove(tbegin, tend, overlaps=True)
tier.Add(ann)
start = end + 1
end += 1
if anchors.GetSize() <= end + 1:
# the last chunk found
if start <= end:
idxstart = anchors[start].GetLabel().GetTypedValue()
idxcur = anchors[end].GetLabel().GetTypedValue()
chunk_text = " ".join(toklist[idxstart:idxcur + 1])
tbegin = anchors[start].GetLocation().GetBegin()
tend = anchors[end].GetLocation().GetEnd()
ann = Annotation(TimeInterval(tbegin, tend),
Label(chunk_text))
tier.Remove(tbegin, tend, overlaps=True)
tier.Add(ann)
to_continue = False
# Begin of the tier
fi = anchors[0].GetLabel().GetTypedValue()
ft = tier[0].GetLocation().GetBegin()
at = anchors[0].GetLocation().GetBegin()
if ft == 0.:
# a silence to start
ft = tier[0].GetLocation().GetEnd()
else:
ft = TimePoint(0.)
if fi > 0 and ft < at:
chunk_text = " ".join(toklist[0:fi])
ann = Annotation(TimeInterval(ft, at), Label(chunk_text))
tier.Add(ann)
# End of the tier
fi = anchors[-1].GetLabel().GetTypedValue()
fi += 1
ft = tier[-1].GetLocation().GetEnd()
at = anchors[-1].GetLocation().GetEnd()
if ft == self._duration:
# a silence to end
ft = tier[-1].GetLocation().GetBegin()
else:
ft = TimePoint(self._duration)
if fi < len(toklist) and at < ft:
chunk_text = " ".join(toklist[fi + 1:len(toklist)])
ann = Annotation(TimeInterval(at, ft), Label(chunk_text))
tier.Add(ann)
chunk_tier = Tier("Chunks")
i = 1
while i < tier.GetSize():
prevtext = tier[i - 1].GetLabel().GetValue()
curtext = tier[i].GetLabel().GetValue()
newtext = prevtext + " " + curtext
if tier[i-1].GetLabel().IsSilence() is False and \
tier[i].GetLabel().IsSilence() is False and \
(len(prevtext.split()) < 3 or len(newtext.split()) < 12):
a = Annotation(
TimeInterval(tier[i - 1].GetLocation().GetBegin(),
tier[i].GetLocation().GetEnd()),
Label(newtext))
chunk_tier.Append(a)
i += 1
else:
chunk_tier.Append(tier[i - 1])
i += 1
return chunk_tier