def test_trim_to_unsupervised_segments():
cut_set = CutSet.from_cuts([
# Yields 3 unsupervised cuts - before first supervision,
# between sup2 and sup3, and after sup3.
Cut('cut1', start=0, duration=30, channel=0, supervisions=[
SupervisionSegment('sup1', 'rec1', start=1.5, duration=8.5),
SupervisionSegment('sup2', 'rec1', start=10, duration=5),
SupervisionSegment('sup3', 'rec1', start=20, duration=8),
]),
# Does not yield any "unsupervised" cut.
Cut('cut2', start=0, duration=30, channel=0, supervisions=[
SupervisionSegment('sup4', 'rec1', start=0, duration=30),
]),
])
unsupervised_cuts = cut_set.trim_to_unsupervised_segments()
assert len(unsupervised_cuts) == 3
assert unsupervised_cuts[0].start == 0
assert unsupervised_cuts[0].duration == 1.5
assert unsupervised_cuts[0].supervisions == []
assert unsupervised_cuts[1].start == 15
assert unsupervised_cuts[1].duration == 5
assert unsupervised_cuts[1].supervisions == []
assert unsupervised_cuts[2].start == 28
assert unsupervised_cuts[2].duration == 2
assert unsupervised_cuts[2].supervisions == []
def test_mix_same_recording_channels():
recording = Recording('rec',
sampling_rate=8000,
num_samples=30 * 8000,
duration=30,
sources=[
AudioSource('file',
channels=[0],
source='irrelevant1.wav'),
AudioSource('file',
channels=[1],
source='irrelevant2.wav')
])
cut_set = CutSet.from_cuts([
Cut('cut1', start=0, duration=30, channel=0, recording=recording),
Cut('cut2', start=0, duration=30, channel=1, recording=recording)
])
mixed = cut_set.mix_same_recording_channels()
assert len(mixed) == 1
cut = mixed[0]
assert isinstance(cut, MixedCut)
assert len(cut.tracks) == 2
assert cut.tracks[0].cut == cut_set[0]
assert cut.tracks[1].cut == cut_set[1]
def test_trim_to_supervisions_mixed_cuts():
cut_set = CutSet.from_cuts([
Cut('cut1',
start=0,
duration=30,
channel=0,
recording=Recording(id='rec1',
sources=[],
sampling_rate=16000,
num_samples=160000,
duration=10.0),
supervisions=[
SupervisionSegment('sup1', 'rec1', start=1.5, duration=8.5),
SupervisionSegment('sup2', 'rec1', start=10, duration=5),
SupervisionSegment('sup3', 'rec1', start=20, duration=8),
]).append(
Cut('cut2',
start=0,
duration=30,
channel=0,
recording=Recording(id='rec1',
sources=[],
sampling_rate=16000,
num_samples=160000,
duration=10.0),
supervisions=[
SupervisionSegment('sup4',
'rec1',
start=0,
duration=30),
]))
])
assert isinstance(cut_set[0], MixedCut)
cuts = cut_set.trim_to_supervisions()
assert len(cuts) == 4
# After "trimming", the MixedCut "decayed" into simple, unmixed cuts, as they did not overlap
assert all(isinstance(cut, Cut) for cut in cuts)
assert all(len(cut.supervisions) == 1 for cut in cuts)
assert all(cut.supervisions[0].start == 0 for cut in cuts)
cut = cuts[0]
# Check that the cuts preserved their start/duration/supervisions after trimming
assert cut.start == 1.5
assert cut.duration == 8.5
assert cut.supervisions[0].id == 'sup1'
cut = cuts[1]
assert cut.start == 10
assert cut.duration == 5
assert cut.supervisions[0].id == 'sup2'
cut = cuts[2]
assert cut.start == 20
assert cut.duration == 8
assert cut.supervisions[0].id == 'sup3'
cut = cuts[3]
assert cut.start == 0
assert cut.duration == 30
assert cut.supervisions[0].id == 'sup4'
def libri_cut():
return Cut(
channel=0,
duration=16.04,
features=Features(
channels=0,
duration=16.04,
num_features=40,
num_frames=1604,
frame_shift=0.01,
recording_id='recording-1',
sampling_rate=16000,
start=0.0,
storage_path='test/fixtures/libri/storage',
storage_key='30c2440c-93cb-4e83-b382-f2a59b3859b4.llc',
storage_type='lilcom_files',
type='fbank',
),
recording=Recording(
id='recording-1',
sources=[
AudioSource(
type='file',
channels=[0],
source='test/fixtures/libri/libri-1088-134315-0000.wav',
)],
sampling_rate=16000,
num_samples=256640,
duration=1604,
),
id='849e13d8-61a2-4d09-a542-dac1aee1b544',
start=0.0,
supervisions=[],
)
def libri_cut():
return Cut(
channel=0,
duration=16.04,
features=Features(
channels=0,
duration=16.04,
num_features=40,
num_frames=1604,
recording_id='recording-1',
sampling_rate=16000,
start=0.0,
storage_path=
'test/fixtures/libri/storage/fc37eb69-43a8-4e6f-a302-646a76606b38.llc',
storage_type='lilcom',
type='fbank',
),
recording=Recording(
id='recording-1',
sources=[
AudioSource(
type='file',
channels=[0],
source='test/fixtures/libri/libri-1088-134315-0000.wav',
)
],
sampling_rate=16000,
num_samples=256640,
duration=1604,
),
id='849e13d8-61a2-4d09-a542-dac1aee1b544',
start=0.0,
supervisions=[],
)
def overlapping_supervisions_cut():
return Cut(
id='cut-1',
start=0.0,
duration=0.5,
channel=0,
features=Features(
recording_id='recording-1',
channels=0,
start=0,
duration=0.5,
type='fbank',
num_frames=50,
num_features=80,
frame_shift=0.01,
sampling_rate=16000,
storage_type='lilcom',
storage_path='test/fixtures/dummy_feats/storage/',
storage_key='e66b6386-aee5-4a5a-8369-fdde1d2b97c7.llc'
),
supervisions=[
SupervisionSegment(id='s1', recording_id='recording-1', start=0.0, duration=0.2),
SupervisionSegment(id='s2', recording_id='recording-1', start=0.1, duration=0.2),
SupervisionSegment(id='s3', recording_id='recording-1', start=0.2, duration=0.2),
SupervisionSegment(id='s4', recording_id='recording-1', start=0.3, duration=0.2)
]
)
def cut_with_relative_paths():
return Cut('cut', 0, 10, 0,
features=Features(type='fbank', num_frames=1000, num_features=40, sampling_rate=8000,
storage_type='lilcom_files', storage_path='storage_dir', storage_key='feats.llc',
start=0,
duration=10),
recording=Recording('rec', [AudioSource('file', [0], 'audio.wav')], 8000, 80000, 10.0)
)
def cut2(dummy_features):
return Cut(id='cut-2',
start=180.0,
duration=10.0,
features=dummy_features,
supervisions=[
SupervisionSegment(id='sup-3',
recording_id='irrelevant',
start=3.0,
duration=2.5)
])
def dummy_cut(id: str = 'irrelevant',
start: float = 0.0,
duration: float = 1.0,
supervisions=None):
return Cut(
id=id,
start=start,
duration=duration,
features=dummy_features(0),
supervisions=supervisions if supervisions is not None else [],
)
def dummy_cut(unique_id: int,
start: float = 0.0,
duration: float = 1.0,
supervisions=None):
return Cut(
id=f'dummy-cut-{unique_id:04d}',
start=start,
duration=duration,
channel=0,
features=dummy_features(unique_id),
supervisions=supervisions if supervisions is not None else [],
)
def test_trim_to_supervisions_mixed_cuts():
cut_set = CutSet.from_cuts([
Cut('cut1',
start=0,
duration=30,
channel=0,
supervisions=[
SupervisionSegment('sup1', 'rec1', start=1.5, duration=8.5),
SupervisionSegment('sup2', 'rec1', start=10, duration=5),
SupervisionSegment('sup3', 'rec1', start=20, duration=8),
]).append(
Cut('cut2',
start=0,
duration=30,
channel=0,
supervisions=[
SupervisionSegment('sup4',
'rec1',
start=0,
duration=30),
]))
])
cuts = cut_set.trim_to_supervisions()
assert len(cuts) == 4
assert all(isinstance(cut, MixedCut) for cut in cuts)
assert all(cut.start == 0 for cut in cuts)
assert all(len(cut.supervisions) == 1 for cut in cuts)
assert all(cut.supervisions[0].start == 0 for cut in cuts)
cut = cuts[0]
assert cut.duration == 8.5
assert cut.supervisions[0].id == 'sup1'
cut = cuts[1]
assert cut.duration == 5
assert cut.supervisions[0].id == 'sup2'
cut = cuts[2]
assert cut.duration == 8
assert cut.supervisions[0].id == 'sup3'
cut = cuts[3]
assert cut.duration == 30
assert cut.supervisions[0].id == 'sup4'
def cut1(dummy_features):
return Cut(id='cut-1',
start=0.0,
duration=10.0,
features=dummy_features,
supervisions=[
SupervisionSegment(id='sup-1',
recording_id='irrelevant',
start=0.5,
duration=6.0),
SupervisionSegment(id='sup-2',
recording_id='irrelevant',
start=7.0,
duration=2.0)
])
