def test_match(self):
utt_filter = subview.MatchingUtteranceIdxFilter(
utterance_idxs={'a', 'b', 'd'})
assert utt_filter.match(tracks.Utterance('a', 'x'), None)
assert utt_filter.match(tracks.Utterance('b', 'x'), None)
assert utt_filter.match(tracks.Utterance('d', 'x'), None)
assert not utt_filter.match(tracks.Utterance('c', 'x'), None)
assert not utt_filter.match(tracks.Utterance('e', 'x'), None)
def test_match_inverse(self):
filter = subview.MatchingUtteranceIdxFilter(
utterance_idxs={'a', 'b', 'd'}, inverse=True)
assert not filter.match(tracks.Utterance('a', 'x'), None)
assert not filter.match(tracks.Utterance('b', 'x'), None)
assert not filter.match(tracks.Utterance('d', 'x'), None)
assert filter.match(tracks.Utterance('c', 'x'), None)
assert filter.match(tracks.Utterance('e', 'x'), None)
def test_read_samples(self):
path = resources.sample_wav_file('wav_1.wav')
track = tracks.FileTrack('wav', path)
issuer = issuers.Issuer('toni')
utt = tracks.Utterance('t', track, issuer=issuer, start=1.0, end=2.30)
l1 = annotations.Label('a', 0.15, 0.448)
l2 = annotations.Label('a', 0.5, 0.73)
ll = annotations.LabelList(labels=[l1, l2])
utt.set_label_list(ll)
expected, __ = librosa.core.load(path,
sr=None,
offset=1.15,
duration=0.298)
assert np.array_equal(l1.read_samples(), expected)
expected, __ = librosa.core.load(path,
sr=None,
offset=1.5,
duration=1.73 - 1.5)
print(expected.shape)
print(l2.read_samples().shape)
assert np.array_equal(l2.read_samples(), expected)
def generate_utterances(track, issuer, n, n_ll_range, n_label_range, rand=None):
if rand is None:
rand = random.Random()
items = []
for i in range(n):
utt_idx = '{}-utt-{}'.format(track.idx, i)
start = rand.random() * 3
end = 3 + rand.random() * 8
utt = tracks.Utterance(
utt_idx,
track,
issuer=issuer,
start=start,
end=end
)
n_ll = rand.randint(*n_ll_range)
for ll in generate_label_lists(n_ll, n_label_range, rand=rand):
utt.set_label_list(ll)
items.append(utt)
return items
def test_split_with_cutting_point_after_end_returns_one_utt(self):
utt = tracks.Utterance('utt-1', None, start=4.0, end=20.0)
res = utt.split([24.5])
assert len(res) == 1
assert res[0].start == 4.0
assert res[0].end == 20.0
def test_does_utt_match_target_format_with_invalid_format_returns_false(self):
file_path = resources.get_resource_path(('audio_formats', 'mp3_2_44_1k_16b.mp3'))
track = tracks.FileTrack('t', file_path)
utt = tracks.Utterance('u', track)
c = conversion.WavAudioFileConverter()
assert not c._does_utt_match_target_format(utt)
def test_does_utt_match_target_format_returns_true(self):
file_path = resources.sample_wav_file('wav_1.wav')
track = tracks.FileTrack('t', file_path)
utt = tracks.Utterance('u', track)
c = conversion.WavAudioFileConverter()
assert c._does_utt_match_target_format(utt)
def test_split_utt_relative_with_labels(self):
ll_1 = annotations.LabelList('phones', labels=[
annotations.Label('alpha', start=0.0, end=30.0)
])
ll_2 = annotations.LabelList('words', labels=[
annotations.Label('b', start=8.0, end=30.0)
])
utt = tracks.Utterance('utt-1', 'file-x', start=10.0, end=40.0, label_lists=[ll_1, ll_2])
res = utt.split([14.0], track_relative=False)
assert len(res) == 2
assert res[0].start == 10.0
assert res[0].end == 24.0
assert res[0].label_lists['phones'] == annotations.LabelList(idx='phones', labels=[
annotations.Label('alpha', 0.0, 14.0)
])
assert res[0].label_lists['words'] == annotations.LabelList(idx='words', labels=[
annotations.Label('b', 8.0, 14.0)
])
assert res[1].start == 24.0
assert res[1].end == 40.0
assert res[1].label_lists['phones'] == annotations.LabelList(idx='phones', labels=[
annotations.Label('alpha', 0.0, 16.0)
])
assert res[1].label_lists['words'] == annotations.LabelList(idx='words', labels=[
annotations.Label('b', 0.0, 16.0)
])
def test_split(self):
ll_1 = annotations.LabelList('phones', labels=[
annotations.Label('alpha', start=0.0, end=30.0)
])
ll_2 = annotations.LabelList('words', labels=[
annotations.Label('b', start=0.0, end=30.0)
])
utt = tracks.Utterance('utt-1', 'track-x', start=0.0, end=40.0, label_lists=[ll_1, ll_2])
res = utt.split([14.0, 29.5])
assert len(res) == 3
assert res[0].start == 0.0
assert res[0].end == 14.0
assert 'phones' in res[0].label_lists.keys()
assert 'words' in res[0].label_lists.keys()
assert res[1].start == 14.0
assert res[1].end == 29.5
assert 'phones' in res[1].label_lists.keys()
assert 'words' in res[1].label_lists.keys()
assert res[2].start == 29.5
assert res[2].end == 40.0
assert 'phones' in res[2].label_lists.keys()
assert 'words' in res[2].label_lists.keys()
def test_encode_label_ends_at_utterance_end(self):
track = tracks.FileTrack('file1',
resources.sample_wav_file('med_len.wav'))
utt = tracks.Utterance('utt1', track, start=3, end=14)
ll = annotations.LabelList(labels=[
annotations.Label('speech', 0, 4),
annotations.Label('music', 4, 9),
annotations.Label('speech', 9, float('inf')),
])
utt.set_label_list(ll)
enc = encoding.FrameHotEncoder(['music', 'speech', 'noise'],
'default',
frame_settings=units.FrameSettings(
32000, 16000),
sr=16000)
actual = enc.encode_utterance(utt)
expected = np.array([
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[1, 1, 0],
[1, 0, 0],
[1, 0, 0],
[1, 0, 0],
[1, 0, 0],
[1, 1, 0],
[0, 1, 0],
]).astype(np.float32)
assert np.array_equal(expected, actual)
def test_import_utterance_no_track(self, corpus):
importing_utterances = [
tracks.Utterance('a', tracks.FileTrack('notexist', 'notexist'),
corpus.issuers['existing_issuer'], 0, 10)
]
with pytest.raises(ValueError):
corpus.import_utterances(importing_utterances)
def test_split_sets_track(self):
file = tracks.FileTrack('file-1', '/some/path')
utt = tracks.Utterance('utt-1', file, start=0.0, end=10.0)
res = utt.split([5.2])
assert len(res) == 2
assert res[0].track == file
assert res[1].track == file
def test_import_utterance_no_issuer(self, corpus):
importing_utterances = [
tracks.Utterance('a', corpus.tracks['existing_file'],
issuers.Issuer('notexist'), 0, 10)
]
with pytest.raises(ValueError):
corpus.import_utterances(importing_utterances)
def test_split_sets_issuer(self):
issuer = issuers.Speaker('spk-1')
utt = tracks.Utterance('utt-1', None, issuer=issuer, start=0.0, end=10.0)
res = utt.split([5.2])
assert len(res) == 2
assert res[0].issuer == issuer
assert res[1].issuer == issuer
def test_encode_utterance_with_single_label(self):
ll = annotations.LabelList(idx='go',
labels=[annotations.Label('a c b')])
utt = tracks.Utterance('utt-1', None, label_lists=ll)
encoder = encoding.TokenOrdinalEncoder('go', ['a', 'b', 'c'])
encoded = encoder.encode_utterance(utt)
assert np.array_equal(encoded, [0, 2, 1])
def test_split_when_utt_start_is_not_zero(self):
utt = tracks.Utterance('utt-1', None, start=6.0, end=20.0)
res = utt.split([3.0])
assert len(res) == 2
assert res[0].start == 6.0
assert res[0].end == 9.0
assert res[1].start == 9.0
assert res[1].end == 20.
def test_split_utt_relative(self):
utt = tracks.Utterance('utt-1', None, start=6.0, end=20.0)
res = utt.split([8.0], track_relative=False)
assert len(res) == 2
assert res[0].start == 6.0
assert res[0].end == 14.0
assert res[1].start == 14.0
assert res[1].end == 20.0
def test_split_endless(self):
utt = tracks.Utterance('utt-1', None, start=0.0)
res = utt.split([24.5])
assert len(res) == 2
assert res[0].start == 0.0
assert res[0].end == 24.5
assert res[1].start == 24.5
assert res[1].end == float('inf')
def test_encode_utterance_with_non_existing_label_list_raises_error(self):
ll = annotations.LabelList(idx='go',
labels=[annotations.Label('a c b unknown')])
utt = tracks.Utterance('utt-1', None, label_lists=ll)
encoder = encoding.TokenOrdinalEncoder('not_existing', ['a', 'b', 'c'])
with pytest.raises(ValueError):
encoder.encode_utterance(utt)
def new_utterance(self,
utterance_idx,
track_idx,
issuer_idx=None,
start=0,
end=float('inf')):
"""
Add a new utterance to the corpus with the given data.
Parameters:
track_idx (str): The track id the utterance is in.
utterance_idx (str): The id to associate with the utterance.
If None or already exists, one is generated.
issuer_idx (str): The issuer id to associate with the utterance.
start (float): Start of the utterance within the track [seconds].
end (float): End of the utterance within the track [seconds].
``inf`` equals the end of the track.
Returns:
Utterance: The newly added utterance.
"""
new_utt_idx = utterance_idx
# Check if there is a track with the given idx
if track_idx not in self._tracks.keys():
raise ValueError(
'Track with id {} does not exist!'.format(track_idx))
# Check if issuer exists
issuer = None
if issuer_idx is not None:
if issuer_idx not in self._issuers.keys():
raise ValueError(
'Issuer with id {} does not exist!'.format(issuer_idx))
else:
issuer = self._issuers[issuer_idx]
# Add index to idx if already existing
if new_utt_idx in self._utterances.keys():
new_utt_idx = naming.index_name_if_in_list(new_utt_idx,
self._utterances.keys())
new_utt = tracks.Utterance(new_utt_idx,
self.tracks[track_idx],
issuer=issuer,
start=start,
end=end)
self._utterances[new_utt_idx] = new_utt
return new_utt
def test_import_utterances(self, corpus):
importing_utterances = [
tracks.Utterance('a', corpus.tracks['existing_file'],
corpus.issuers['existing_issuer'], 0, 10),
tracks.Utterance('b', corpus.tracks['existing_file'],
corpus.issuers['existing_issuer'], 10, 20),
tracks.Utterance('existing_utt', corpus.tracks['existing_file'],
corpus.issuers['existing_issuer'], 20, 30)
]
mapping = corpus.import_utterances(importing_utterances)
assert corpus.num_utterances == 4
assert 'a' in corpus.utterances.keys()
assert 'b' in corpus.utterances.keys()
assert 'existing_utt_1' in corpus.utterances.keys()
assert len(mapping) == 3
assert mapping['a'].idx == 'a'
assert mapping['b'].idx == 'b'
assert mapping['existing_utt'].idx == 'existing_utt_1'
def corpus():
corpus = audiomate.Corpus()
ex_file = tracks.FileTrack('existing_file', '../any/path.wav')
ex_issuer = issuers.Issuer('existing_issuer')
ex_utterance = tracks.Utterance('existing_utt', ex_file, issuer=ex_issuer)
corpus.tracks['existing_file'] = ex_file
corpus.issuers['existing_issuer'] = ex_issuer
corpus.utterances['existing_utt'] = ex_utterance
return corpus
def test_split_with_overlap(self):
ll_1 = annotations.LabelList('phones',
labels=[
annotations.Label('alpha',
start=0.0,
end=30.0),
annotations.Label('bravo',
start=20.0,
end=42.0)
])
ll_2 = annotations.LabelList(
'words', labels=[annotations.Label('b', start=8.0, end=30.0)])
utt = tracks.Utterance('utt-1',
'file-x',
start=10.0,
end=55.0,
label_lists=[ll_1, ll_2])
res = utt.split([12.0, 24.0], track_relative=False, overlap=2.0)
assert len(res) == 3
assert res[0].start == 10.0
assert res[0].end == 24.0
assert res[0].label_lists['phones'] == annotations.LabelList(
idx='phones', labels=[annotations.Label('alpha', 0.0, 14.0)])
assert res[0].label_lists['words'] == annotations.LabelList(
idx='words', labels=[annotations.Label('b', 8.0, 14.0)])
print(res[1].label_lists['phones'].labels)
assert res[1].start == 20.0
assert res[1].end == 36.0
assert res[1].label_lists['phones'] == annotations.LabelList(
idx='phones',
labels=[
annotations.Label('alpha', 0.0, 16.0),
annotations.Label('bravo', 10.0, 16.0),
])
assert res[1].label_lists['words'] == annotations.LabelList(
idx='words', labels=[annotations.Label('b', 0.0, 16.0)])
assert res[2].start == 32.0
assert res[2].end == 55.0
assert res[2].label_lists['phones'] == annotations.LabelList(
idx='phones',
labels=[
annotations.Label('alpha', 0.0, 8.0),
annotations.Label('bravo', 0.0, 20.0),
])
assert res[2].label_lists['words'] == annotations.LabelList(
idx='words', labels=[annotations.Label('b', 0.0, 8.0)])
def setup_method(self):
self.ll_1 = annotations.LabelList(idx='alpha',
labels=[
annotations.Label('a', 3.2, 4.5),
annotations.Label('b', 5.1, 8.9),
annotations.Label(
'c', 7.2, 10.5),
annotations.Label('d', 10.5, 14),
annotations.Label('d', 15, 18)
])
self.ll_2 = annotations.LabelList(idx='bravo',
labels=[
annotations.Label('a', 1.0, 4.2),
annotations.Label('e', 4.2, 7.9),
annotations.Label(
'c', 7.2, 10.5),
annotations.Label('f', 10.5, 14),
annotations.Label('d', 15, 17.3)
])
self.ll_duplicate_idx = annotations.LabelList(
idx='charlie',
labels=[
annotations.Label('t', 1.0, 4.2),
annotations.Label('h', 4.2, 7.9)
])
self.ll_3 = annotations.LabelList(idx='charlie',
labels=[
annotations.Label('a', 1.0, 4.2),
annotations.Label('g', 4.2, 7.9)
])
self.track = tracks.FileTrack('wav',
resources.sample_wav_file('wav_1.wav'))
self.issuer = issuers.Issuer('toni')
self.utt = tracks.Utterance('test',
self.track,
issuer=self.issuer,
start=1.25,
end=1.30,
label_lists=[
self.ll_1, self.ll_2,
self.ll_duplicate_idx, self.ll_3
])
def test_utt_read_samples(benchmark):
utts = []
wav_path = resources.get_test_resource_path(('wav_files', 'med_len.wav'))
track = tracks.FileTrack('idx', wav_path)
utts.append(tracks.Utterance('uidx', track))
utts.append(tracks.Utterance('uidx', track, start=2.8))
utts.append(tracks.Utterance('uidx', track, end=10.2))
utts.append(tracks.Utterance('uidx', track, start=2.4, end=9.8))
mp3_path = resources.get_test_resource_path(
('audio_formats', 'mp3_2_44_1k_16b.mp3'))
track = tracks.FileTrack('idx', mp3_path)
utts.append(tracks.Utterance('uidx', track))
utts.append(tracks.Utterance('uidx', track, start=2.8))
utts.append(tracks.Utterance('uidx', track, end=4.9))
utts.append(tracks.Utterance('uidx', track, start=0.4, end=4.8))
benchmark(run, utts)
def test_encode_utterance_with_multiple_non_overlapping_labels(self):
ll = annotations.LabelList(idx='go',
labels=[
annotations.Label('a c b',
start=0,
end=5),
annotations.Label('c b b',
start=5,
end=9.4),
annotations.Label('a a a',
start=9.5,
end=10.2)
])
utt = tracks.Utterance('utt-1', None, label_lists=ll)
encoder = encoding.TokenOrdinalEncoder('go', ['a', 'b', 'c'])
encoded = encoder.encode_utterance(utt)
assert np.array_equal(encoded, [0, 2, 1, 2, 1, 1, 0, 0, 0])
def utt_with_noise():
utt = tracks.Utterance('utt-2', 'file-2')
utt.set_label_list(
annotations.LabelList(idx='alpha',
labels=[
annotations.Label('music', 0, 5),
annotations.Label('speech', 5, 12),
annotations.Label('noise', 13, 15)
]))
utt.set_label_list(
annotations.LabelList(idx='bravo',
labels=[
annotations.Label('music', 0, 1),
annotations.Label('speech', 2, 6)
]))
return utt
def test_encode_utterance_with_overlapping_labels_raises_error(self):
ll = annotations.LabelList(idx='go',
labels=[
annotations.Label('a c b',
start=0,
end=5),
annotations.Label('c b b',
start=2,
end=9.4),
annotations.Label('a a a',
start=9.5,
end=10.2)
])
utt = tracks.Utterance('utt-1', None, label_lists=ll)
encoder = encoding.TokenOrdinalEncoder('go', ['a', 'b', 'c'])
with pytest.raises(ValueError):
encoder.encode_utterance(utt)
def test_encode_utterance_takes_lower_index_first(self):
file = tracks.FileTrack('file-idx',
resources.sample_wav_file('wav_1.wav'))
utt = tracks.Utterance('utt-idx', file, start=0, end=5)
ll = annotations.LabelList(labels=[
annotations.Label('music', 0, 3),
annotations.Label('speech', 3, 5)
])
utt.set_label_list(ll)
enc = encoding.FrameOrdinalEncoder(['speech', 'music', 'noise'],
'default',
frame_settings=units.FrameSettings(
32000, 16000),
sr=16000)
actual = enc.encode_utterance(utt)
expected = np.array([1, 1, 0, 0]).astype(np.int)
assert np.array_equal(expected, actual)
def test_validate_utterance_returns_completly_outlying_label(self):
utt = tracks.Utterance('utt-idx', None, start=10.0, end=17.9)
ll = annotations.LabelList(idx='default',
labels=[
annotations.Label('a',
start=-4.0,
end=-2.0),
annotations.Label('b',
start=19.0,
end=22.0),
])
utt.set_label_list(ll)
val = validation.LabelOverflowValidator('default')
result = val.validate_utterance(utt)
result = sorted(result, key=lambda x: x[0])
assert len(result) == 2
assert result[0] == (-4.0, -2.0, 'a')
assert result[1] == (19.0, 22.0, 'b')
