def test_relabel_flattens_full_overlap_into_combined_label(self):
projections = {
('a', ): 'a',
('b', ): 'b',
('a', 'b'): 'a_b',
}
label_list = assets.LabelList(
labels=[assets.Label('a', 3.2, 4.9),
assets.Label('b', 3.9, 4.5)])
actual = relabeling.relabel(label_list, projections)
assert len(actual) == 3
assert actual[0].start == 3.2
assert actual[0].end == 3.9
assert actual[0].value == 'a'
assert actual[1].start == 3.9
assert actual[1].end == 4.5
assert actual[1].value == 'a_b'
assert actual[2].start == 4.5
assert actual[2].end == 4.9
assert actual[2].value == 'a'
def setUp(self):
self.utt1 = assets.Utterance('utt-1', 'file-1')
self.utt1.set_label_list(
assets.LabelList(idx='alpha',
labels=[
assets.Label('music', 0, 5),
assets.Label('speech', 5, 12),
assets.Label('music', 13, 15)
]))
self.utt1.set_label_list(
assets.LabelList(idx='bravo',
labels=[
assets.Label('music', 0, 1),
assets.Label('speech', 2, 6)
]))
self.utt2 = assets.Utterance('utt-2', 'file-2')
self.utt2.set_label_list(
assets.LabelList(idx='alpha',
labels=[
assets.Label('music', 0, 5),
assets.Label('speech', 5, 12),
assets.Label('noise', 13, 15)
]))
self.utt2.set_label_list(
assets.LabelList(idx='bravo',
labels=[
assets.Label('music', 0, 1),
assets.Label('speech', 2, 6)
]))
def test_relabel_proceeds_despite_unmapped_labels_in_presence_of_wildcard_rule(
self):
label_list = assets.LabelList(labels=[
assets.Label('a', 3.2, 5.1),
assets.Label('b', 4.2, 4.7),
assets.Label('c', 4.3, 4.8)
])
actual = relabeling.relabel(label_list, {
('a', ): 'new_label_a',
('**', ): 'catch_all'
})
assert len(actual) == 5
assert actual[0].start == 3.2
assert actual[0].end == 4.2
assert actual[0].value == 'new_label_a'
assert actual[1].start == 4.2
assert actual[1].end == 4.3
assert actual[1].value == 'catch_all'
assert actual[2].start == 4.3
assert actual[2].end == 4.7
assert actual[2].value == 'catch_all'
assert actual[3].start == 4.7
assert actual[3].end == 4.8
assert actual[3].value == 'catch_all'
assert actual[4].start == 4.8
assert actual[4].end == 5.1
assert actual[4].value == 'new_label_a'
def test_read_samples(self):
file = assets.File('wav', resources.sample_wav_file('wav_1.wav'))
issuer = assets.Issuer('toni')
utt = assets.Utterance('test',
file,
issuer=issuer,
start=1.0,
end=2.30)
l1 = assets.Label('a', 0.15, 0.448)
l2 = assets.Label('a', 0.5, 0.73)
ll = assets.LabelList(labels=[l1, l2])
utt.set_label_list(ll)
expected, __ = librosa.core.load(file.path,
sr=None,
offset=1.15,
duration=0.298)
assert np.array_equal(l1.read_samples(), expected)
expected, __ = librosa.core.load(file.path,
sr=None,
offset=1.5,
duration=0.23)
assert np.array_equal(l2.read_samples(), expected)
def test_all_projections_missing_if_no_projections_defined(self):
label_list = assets.LabelList(labels=[
assets.Label('b', 3.2, 4.5),
assets.Label('a', 4.0, 4.9),
assets.Label('c', 4.2, 5.1)
])
unmapped_combinations = relabeling.find_missing_projections(
label_list, {})
assert len(unmapped_combinations) == 5
assert ('b', ) in unmapped_combinations
assert (
'a',
'b',
) in unmapped_combinations
assert (
'a',
'b',
'c',
) in unmapped_combinations
assert (
'a',
'c',
) in unmapped_combinations
assert ('c', ) in unmapped_combinations
def test_ranges_zero_to_end(self):
ll = assets.LabelList(
labels=[assets.Label('a', 0, -1),
assets.Label('b', 5.1, 8.9)])
ranges = ll.ranges()
r = next(ranges)
self.assertEqual(0, r[0])
self.assertEqual(5.1, r[1])
self.assertIn(ll[0], r[2])
r = next(ranges)
self.assertEqual(5.1, r[0])
self.assertEqual(8.9, r[1])
self.assertIn(ll[0], r[2])
self.assertIn(ll[1], r[2])
r = next(ranges)
self.assertEqual(8.9, r[0])
self.assertEqual(-1, r[1])
self.assertIn(ll[0], r[2])
with self.assertRaises(StopIteration):
next(ranges)
def test_relabel_removes_overlapping_segment(self):
projections = {
('a', ): 'a',
(
'a',
'b',
): '',
('b', ): 'b',
}
label_list = assets.LabelList(
labels=[assets.Label('a', 3.2, 5.1),
assets.Label('b', 4.2, 4.7)])
actual = relabeling.relabel(label_list, projections)
assert len(actual) == 2
assert actual[0].start == 3.2
assert actual[0].end == 4.2
assert actual[0].value == 'a'
assert actual[1].start == 4.7
assert actual[1].end == 5.1
assert actual[1].value == 'a'
def test_no_missing_projections_if_projection_complete(self):
projections = {
('b', ): 'foo',
(
'a',
'b',
): 'foo',
(
'a',
'b',
'c',
): 'foo',
(
'a',
'c',
): 'foo',
('c', ): 'bar'
}
label_list = assets.LabelList(labels=[
assets.Label('b', 3.2, 4.5),
assets.Label('a', 4.0, 4.9),
assets.Label('c', 4.2, 5.1)
])
unmapped_combinations = relabeling.find_missing_projections(
label_list, projections)
assert len(unmapped_combinations) == 0
def test_eq_ignores_label_list_relation(self):
a = assets.Label('some label A', 1.0, 2.0)
b = assets.Label('some label a', 1.0, 2.0)
al = assets.LabelList(idx='one', labels=[a])
bl = assets.LabelList(idx='another', labels=[b])
assert a.label_list == al
assert b.label_list == bl
assert a == b
def test_extend(self):
ll = assets.LabelList()
label_a = assets.Label('some text')
label_b = assets.Label('more text')
label_c = assets.Label('text again')
ll.extend([label_a, label_b, label_c])
assert len(ll) == 3
assert label_a.label_list == ll
assert label_b.label_list == ll
assert label_c.label_list == ll
def test_label_total_durations(self):
ll = assets.LabelList(labels=[
assets.Label('a', 3.2, 4.5),
assets.Label('b', 5.1, 8.9),
assets.Label('c', 7.2, 10.5),
assets.Label('a', 10.5, 14),
assets.Label('c', 13, 14)
])
res = ll.label_total_duration()
assert res['a'] == pytest.approx(4.8)
assert res['b'] == pytest.approx(3.8)
assert res['c'] == pytest.approx(4.3)
def test_label_count(self):
ll = assets.LabelList(labels=[
assets.Label('a', 3.2, 4.5),
assets.Label('b', 5.1, 8.9),
assets.Label('c', 7.2, 10.5),
assets.Label('a', 10.5, 14),
assets.Label('c', 13, 14)
])
res = ll.label_count()
self.assertEqual(2, res['a'])
self.assertEqual(1, res['b'])
self.assertEqual(2, res['c'])
def test_ranges_include_labels(self):
ll = assets.LabelList(
labels=[assets.Label('a', 3.2, 4.5),
assets.Label('b', 5.1, 8.9)])
ranges = ll.ranges(include_labels=['a'])
r = next(ranges)
self.assertEqual(3.2, r[0])
self.assertEqual(4.5, r[1])
self.assertIn(ll[0], r[2])
with self.assertRaises(StopIteration):
next(ranges)
def corpus_with_more_labels():
"""
Corpus with an extra label-list.
"""
corpus = resources.create_single_label_corpus()
corpus.utterances['utt-1'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('alpha')]))
corpus.utterances['utt-2'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('alpha')]))
corpus.utterances['utt-3'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
corpus.utterances['utt-4'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
corpus.utterances['utt-5'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
corpus.utterances['utt-6'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
corpus.utterances['utt-7'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
corpus.utterances['utt-8'].set_label_list(
assets.LabelList(idx='radio', labels=[assets.Label('beta')]))
return corpus
def test_merge_consecutive_labels_with_same_values():
ll = assets.LabelList(labels=[
assets.Label('a', 0, 0.993),
assets.Label('a', 1.001, 2.8),
assets.Label('b', 2.8, 3.94)
])
label_cleaning.merge_consecutive_labels_with_same_values(ll,
threshold=0.01)
assert len(ll) == 2
assert ll[0] == assets.Label('a', 0, 2.8)
assert ll[1] == assets.Label('b', 2.8, 3.94)
def test_relabel_flattens_partial_overlap_into_combined_label(self):
projections = {
('a', ): 'a',
('b', ): 'b',
('c', ): 'c',
(
'a',
'b',
): 'a_b',
(
'a',
'b',
'c',
): 'a_b_c',
(
'b',
'c',
): 'b_c',
}
label_list = assets.LabelList(labels=[
assets.Label('a', 3.2, 4.5),
assets.Label('b', 4.0, 4.9),
assets.Label('c', 4.2, 5.1)
])
actual = relabeling.relabel(label_list, projections)
assert len(actual) == 5
assert actual[0].start == 3.2
assert actual[0].end == 4.0
assert actual[0].value == 'a'
assert actual[1].start == 4.0
assert actual[1].end == 4.2
assert actual[1].value == 'a_b'
assert actual[2].start == 4.2
assert actual[2].end == 4.5
assert actual[2].value == 'a_b_c'
assert actual[3].start == 4.5
assert actual[3].end == 4.9
assert actual[3].value == 'b_c'
assert actual[4].start == 4.9
assert actual[4].end == 5.1
assert actual[4].value == 'c'
def test_relabel_throws_error_if_unmapped_labels_are_detected(self):
label_list = assets.LabelList(labels=[
assets.Label('a', 3.2, 5.1),
assets.Label('b', 4.2, 4.7),
assets.Label('c', 4.3, 4.8)
])
unmapped_combinations = [('a', 'b'), ('a', 'b', 'c'), ('a', 'c')]
expected_message = 'Unmapped combinations: {}'.format(
unmapped_combinations)
with pytest.raises(relabeling.UnmappedLabelsException) as ex:
relabeling.relabel(label_list, {('a', ): 'foo'})
assert ex.value.message == expected_message
def test_label_creation(self):
a = assets.Label('value', 6.2, 8.9)
assert a.value == 'value'
assert a.start == 6.2
assert a.end == 8.9
assert len(a.meta) == 0
def setUp(self):
file = assets.File('wav', resources.sample_wav_file('wav_1.wav'))
utt = assets.Utterance('utt', file, start=0.3, end=-1)
ll = assets.LabelList()
self.test_label = assets.Label('a', start=0.5, end=-1)
ll.append(self.test_label)
utt.set_label_list(ll)
def read_labels(path, corpus):
for label_file in glob.glob(
os.path.join(path, '{}_*.txt'.format(LABEL_FILE_PREFIX))):
file_name = os.path.basename(label_file)
key = file_name[len('{}_'.format(LABEL_FILE_PREFIX)
):len(file_name) - len('.txt')]
utterance_labels = collections.defaultdict(list)
labels = textfile.read_separated_lines_generator(label_file,
separator=' ',
max_columns=4)
for record in labels:
label = record[3]
start = float(record[1])
end = float(record[2])
meta = None
meta_match = META_PATTERN.match(label)
if meta_match is not None:
meta_json = meta_match.group(2)
meta = json.loads(meta_json)
label = meta_match.group(1)
utterance_labels[record[0]].append(
assets.Label(label, start, end, meta=meta))
for utterance_idx, labels in utterance_labels.items():
ll = assets.LabelList(idx=key, labels=labels)
corpus.utterances[utterance_idx].set_label_list(ll)
def test_missing_projections_are_naturally_sorted(self):
label_list = assets.LabelList(labels=[
assets.Label('b', 1.0, 2.0),
assets.Label('a', 1.5, 2.5),
])
unmapped_combinations = relabeling.find_missing_projections(
label_list, {})
assert len(unmapped_combinations) == 3
assert unmapped_combinations[0] == ('a', )
assert unmapped_combinations[1] == (
'a',
'b',
)
assert unmapped_combinations[2] == ('b', )
def test_no_missing_projections_if_covered_by_catch_all_rule(self):
projections = {
('b', ): 'new_label_b',
('**', ): 'new_label_all',
}
label_list = assets.LabelList(labels=[
assets.Label('b', 3.2, 4.5),
assets.Label('a', 4.0, 4.9),
assets.Label('c', 4.2, 5.1)
])
unmapped_combinations = relabeling.find_missing_projections(
label_list, projections)
assert len(unmapped_combinations) == 0
def test_encode_utterance_takes_lower_index_first(self):
file = assets.File('file-idx', resources.sample_wav_file('wav_1.wav'))
utt = assets.Utterance('utt-idx', file, start=0, end=5)
ll = assets.LabelList(
labels=[assets.Label('music', 0, 3),
assets.Label('speech', 3, 5)])
utt.set_label_list(ll)
enc = label_encoding.FrameOrdinalEncoder(
['speech', 'music', 'noise'],
frame_settings=units.FrameSettings(32000, 16000),
sr=16000)
actual = enc.encode(utt)
expected = np.array([1, 1, 0, 0]).astype(np.int)
assert np.array_equal(expected, actual)
def test_relabel_removes_unwanted_labels(self):
projections = {
('a', ): '',
('b', ): 'b',
}
label_list = assets.LabelList(
labels=[assets.Label('a', 3.2, 4.4),
assets.Label('b', 4.4, 5.1)])
actual = relabeling.relabel(label_list, projections)
assert len(actual) == 1
assert actual[0].start == 4.4
assert actual[0].end == 5.1
assert actual[0].value == 'b'
def test_label_creation_with_info(self):
a = assets.Label('value', 6.2, 8.9, meta={'something': 2})
assert a.value == 'value'
assert a.start == 6.2
assert a.end == 8.9
assert len(a.meta) == 1
assert a.meta['something'] == 2
def test_append(self):
ll = assets.LabelList()
label = assets.Label('some text')
ll.append(label)
assert len(ll) == 1
assert label.label_list == ll
def test_set_label_list(self):
ll_4 = assets.LabelList(
idx='delta',
labels=[assets.Label('y', 0.0, 3.3),
assets.Label('t', 3.8, 7.9)])
self.utt.set_label_list(ll_4)
assert len(self.utt.label_lists) == 4
assert self.utt.label_lists['alpha'] == self.ll_1
assert self.utt.label_lists['bravo'] == self.ll_2
assert self.utt.label_lists['charlie'] == self.ll_3
assert self.utt.label_lists['delta'] == ll_4
assert self.ll_1.utterance == self.utt
assert self.ll_2.utterance == self.utt
assert self.ll_3.utterance == self.utt
assert ll_4.utterance == self.utt
def test_relabel_maps_a_onto_b(self):
label_list = assets.LabelList(labels=[assets.Label('a', 3.2, 4.5)])
actual = relabeling.relabel(label_list, {('a', ): 'b'})
assert len(actual) == 1
assert actual[0].start == 3.2
assert actual[0].end == 4.5
assert actual[0].value == 'b'
def test_no_duplicate_missing_projections_reported(self):
label_list = assets.LabelList(labels=[
assets.Label('b', 1.0, 2.0),
assets.Label('a', 1.5, 2.5),
assets.Label('b', 3.0, 4.0),
assets.Label('a', 3.5, 4.5),
])
unmapped_combinations = relabeling.find_missing_projections(
label_list, {})
assert len(unmapped_combinations) == 3
assert ('b', ) in unmapped_combinations
assert (
'a',
'b',
) in unmapped_combinations
assert ('a', ) in unmapped_combinations
def test_ranges_with_empty(self):
ll = assets.LabelList(labels=[
assets.Label('a', 3.2, 4.5),
assets.Label('b', 5.1, 8.9),
assets.Label('c', 7.2, 10.5),
assets.Label('d', 10.5, 14)
])
ranges = ll.ranges(yield_ranges_without_labels=True)
r = next(ranges)
self.assertEqual(3.2, r[0])
self.assertEqual(4.5, r[1])
self.assertIn(ll[0], r[2])
r = next(ranges)
self.assertEqual(4.5, r[0])
self.assertEqual(5.1, r[1])
self.assertEqual(0, len(r[2]))
r = next(ranges)
self.assertEqual(5.1, r[0])
self.assertEqual(7.2, r[1])
self.assertIn(ll[1], r[2])
r = next(ranges)
self.assertEqual(7.2, r[0])
self.assertEqual(8.9, r[1])
self.assertIn(ll[1], r[2])
self.assertIn(ll[2], r[2])
r = next(ranges)
self.assertEqual(8.9, r[0])
self.assertEqual(10.5, r[1])
self.assertIn(ll[2], r[2])
r = next(ranges)
self.assertEqual(10.5, r[0])
self.assertEqual(14, r[1])
self.assertIn(ll[3], r[2])
with self.assertRaises(StopIteration):
next(ranges)
