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_encode_full_utterance(self):
ds = resources.create_multi_label_corpus()
enc = encoding.FrameHotEncoder(['music', 'speech', 'noise'],
'default',
frame_settings=units.FrameSettings(
32000, 16000),
sr=16000)
actual = enc.encode_utterance(ds.utterances['utt-6'])
expected = np.array([
[1, 0, 0],
[1, 0, 0],
[1, 0, 0],
[1, 0, 0],
[1, 1, 0],
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[0, 1, 0],
[1, 0, 0],
[1, 0, 0],
]).astype(np.float32)
assert np.array_equal(expected, actual)
def process_file(self,
file_path,
frame_size=400,
hop_size=160,
sr=None,
start=0,
end=-1,
utterance=None,
corpus=None):
"""
Process the audio-file in **offline** mode, in one go.
Args:
file_path (str): The audio file to process.
frame_size (int): The number of samples per frame.
hop_size (int): The number of samples between two frames.
sr (int): Use the given sampling rate. If None uses the native sampling rate from the underlying data.
start (float): The point within the file in seconds to start processing from.
end (float): The point within the file in seconds to end processing.
utterance (Utterance): The utterance that is associated with this file, if available.
corpus (Corpus): The corpus this file is part of, if available.
Returns:
np.ndarray: The processed features.
"""
frame_settings = units.FrameSettings(frame_size, hop_size)
if end > 0:
samples, sr = librosa.core.load(file_path,
sr=sr,
offset=start,
duration=end - start)
else:
samples, sr = librosa.core.load(file_path, sr=sr, offset=start)
if samples.size <= 0:
raise ValueError('File {} has no samples'.format(file_path))
# Pad with zeros to match frames
num_frames = frame_settings.num_frames(samples.size)
num_pad_samples = (num_frames - 1) * hop_size + frame_size
if num_pad_samples > samples.size:
samples = np.pad(samples, (0, num_pad_samples - samples.size),
mode='constant',
constant_values=0)
# Get sampling-rate if not given
sampling_rate = sr or utterance.sampling_rate
frames = librosa.util.frame(samples,
frame_length=frame_size,
hop_length=hop_size).T
return self.process_frames(frames,
sampling_rate,
0,
last=True,
utterance=utterance,
corpus=corpus)
def test_encode_utterance(self):
ds = resources.create_multi_label_corpus()
enc = label_encoding.FrameOrdinalEncoder(
['music', 'speech', 'noise'],
frame_settings=units.FrameSettings(32000, 16000),
sr=16000)
actual = enc.encode(ds.utterances['utt-6'])
expected = np.array([0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0,
0]).astype(np.int)
assert np.array_equal(expected, actual)
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_encode_utterance_takes_larger_label(self):
file = tracks.FileTrack('file-idx',
resources.sample_wav_file('wav_1.wav'))
utt = tracks.Utterance('utt-idx', file, start=0, end=8)
ll = annotations.LabelList(labels=[
annotations.Label('music', 0, 4.5),
annotations.Label('speech', 4.5, 8)
])
utt.set_label_list(ll)
enc = encoding.FrameOrdinalEncoder(['music', 'speech', 'noise'],
'default',
frame_settings=units.FrameSettings(
32000, 16000),
sr=16000)
actual = enc.encode_utterance(utt)
expected = np.array([0, 0, 0, 0, 1, 1, 1]).astype(np.int)
assert np.array_equal(expected, actual)
def test_time_range_to_frame_range(self, frame_size, hop_size, start_time, end_time, sr, start_index, end_index):
f = units.FrameSettings(frame_size, hop_size)
assert f.time_range_to_frame_range(start_time, end_time, sr) == (start_index, end_index)
def test_frame_to_seconds(self, frame_size, hop_size, frame_index, sr, start, end):
f = units.FrameSettings(frame_size, hop_size)
assert f.frame_to_seconds(frame_index, sr) == (start, end)
def test_sample_to_frame_range(self, frame_size, hop_size, sample_index, start_frame, end_frame):
f = units.FrameSettings(frame_size, hop_size)
assert f.sample_to_frame_range(sample_index) == (start_frame, end_frame)
def test_frame_to_sample(self, frame_size, hop_size, frame_index, start_sample, end_sample):
f = units.FrameSettings(frame_size, hop_size)
assert f.frame_to_sample(frame_index) == (start_sample, end_sample)
def test_num_frames(self, frame_size, hop_size, num_samples, num_frames):
f = units.FrameSettings(frame_size, hop_size)
assert f.num_frames(num_samples) == num_frames
