def test_multiple_actions(self):
texts = [
'The quick brown fox jumps over the lazy dog',
'Zology raku123456 fasdasd asd4123414 1234584'
]
flows = [
naf.Sequential([
naf.Sometimes([nac.RandomCharAug(action="insert"),
nac.RandomCharAug(action="delete")],
aug_p=0.9),
naf.Sequential([
nac.RandomCharAug(action="substitute", aug_char_min=1, aug_char_p=0.6, aug_word_p=0.6)
], name='Sub_Seq')
]),
naf.Sometimes([
naf.Sometimes([nac.RandomCharAug(action="insert"),
nac.RandomCharAug(action="delete")]),
naf.Sequential([nac.OcrAug(), nac.KeyboardAug(aug_char_min=1),
nac.RandomCharAug(action="substitute", aug_char_min=1, aug_char_p=0.6, aug_word_p=0.6)])
], aug_p=0.9)
]
# Since prob may be low and causing do not perform data augmentation. Retry 5 times
for flow in flows:
for text in texts:
at_least_one_not_equal = False
for _ in range(5):
augmented_text = flow.augment(text, n=1)
if text != augmented_text:
at_least_one_not_equal = True
break
self.assertTrue(at_least_one_not_equal)
self.assertLess(0, len(flows))
self.assertLess(0, len(texts))
def test_multi_thread(self):
text = 'The quick brown fox jumps over the lazy dog'
n = 3
flows = [
naf.Sequential([
naf.Sequential([
nac.OcrAug(),
naw.WordEmbsAug(
model_type='word2vec',
model_path=os.environ["MODEL_DIR"] + 'GoogleNews-vectors-negative300.bin')
]),
naf.Sequential([
nac.RandomCharAug(),
]),
naw.ContextualWordEmbsAug(
model_path='xlnet-base-cased', action="substitute",
skip_unknown_word=True, temperature=0.7, device='cpu')
]),
naf.Sometimes([
naf.Sequential([
nac.OcrAug(),
nac.RandomCharAug(),
]),
naf.Sometimes([
naw.WordEmbsAug(model_type='word2vec',
model_path=os.environ["MODEL_DIR"] + 'GoogleNews-vectors-negative300.bin')
], pipeline_p=0.999),
naw.ContextualWordEmbsAug(
model_path='xlnet-base-cased', action="substitute",
skip_unknown_word=True, temperature=0.7, device='cpu')
], pipeline_p=0.9999)
]
for num_thread in [1, 3]:
for flow in flows:
augmented_data = flow.augment(text, n=n, num_thread=num_thread)
self.assertEqual(len(augmented_data), n)
def test_single_action(self):
texts = [
'The quick brown fox jumps over the lazy dog',
'Zology raku123456 fasdasd asd4123414 1234584 s@#'
]
flow = naf.Sequential([nac.RandomCharAug(action=Action.INSERT, min_char=1)])
for text in texts:
augmented_text = flow.augment(text)
self.assertNotEqual(text, augmented_text)
self.assertLess(0, len(text))
self.assertLess(0, len(texts))
def test_n_output_textual(self):
texts = [
'The quick brown fox jumps over the lazy dog',
'Zology raku123456 fasdasd asd4123414 1234584',
'AAAAAAAAAAA AAAAAAAAAAAAAA'
]
flows = [
naf.Sequential(
[nac.RandomCharAug(action="insert"),
naw.RandomWordAug()]),
naf.Sometimes([
nac.RandomCharAug(action="insert"),
nac.RandomCharAug(action="delete")
],
pipeline_p=0.9),
naf.Sequential([
naf.Sequential(
[nac.RandomCharAug(action="insert"),
naw.RandomWordAug()]),
naf.Sometimes([
nac.RandomCharAug(action="insert"),
nac.RandomCharAug(action="delete")
],
pipeline_p=0.9)
])
]
for flow in flows:
for text in texts:
augmented_texts = flow.augment(text, n=3)
self.assertGreater(len(augmented_texts), 1)
for augmented_text in augmented_texts:
self.assertNotEqual(augmented_text, text)
self.assertLess(0, len(flows))
self.assertLess(0, len(texts))
def test_n_output_spectrogram(self):
audio, sampling_rate = librosa.load(self.sample_wav_file)
mel_spectrogram = LoadUtil.load_mel_spectrogram(self.sample_wav_file,
n_mels=128)
#
flows = [
naf.Sequential([
nas.FrequencyMaskingAug(mask_factor=80),
nas.TimeMaskingAug(mask_factor=80)
]),
naf.Sometimes([
nas.FrequencyMaskingAug(mask_factor=80),
nas.TimeMaskingAug(mask_factor=80)
],
pipeline_p=0.9),
naf.Sequential([
naf.Sequential([
nas.FrequencyMaskingAug(mask_factor=80),
nas.TimeMaskingAug(mask_factor=80)
]),
naf.Sometimes([
nas.FrequencyMaskingAug(mask_factor=80),
nas.TimeMaskingAug(mask_factor=80)
],
pipeline_p=0.9)
])
]
for flow in flows:
augmented_mel_spectrograms = flow.augment(mel_spectrogram, n=3)
self.assertGreater(len(augmented_mel_spectrograms), 1)
for augmented_mel_spectrogram in augmented_mel_spectrograms:
self.assertFalse(
np.array_equal(mel_spectrogram, augmented_mel_spectrogram))
self.assertLess(0, len(flows))
def test_audio(self):
# https://freewavesamples.com/yamaha-v50-rock-beat-120-bpm
sample_wav_file = os.path.abspath(os.path.join(
os.path.dirname(__file__), '..', '..', 'data', 'Yamaha-V50-Rock-Beat-120bpm.wav'))
audio, sampling_rate = librosa.load(sample_wav_file)
flow = naf.Sequential([
naa.NoiseAug(),
naa.PitchAug(sampling_rate=sampling_rate, factor=(0.2, 1.5)),
naa.ShiftAug(sampling_rate=sampling_rate, duration=2),
naa.SpeedAug(factor=(1.5, 3))
])
augmented_audio = flow.augment(audio)
self.assertFalse(np.array_equal(audio, augmented_audio))
self.assertTrue(len(audio), len(augmented_audio))
def test_audio(self):
# https://freewavesamples.com/yamaha-v50-rock-beat-120-bpm
sample_wav_file = os.path.join(os.environ.get("TEST_DIR"), 'res',
'audio',
'Yamaha-V50-Rock-Beat-120bpm.wav')
audio, sampling_rate = AudioLoader.load_audio(sample_wav_file)
flow = naf.Sequential([
naa.NoiseAug(),
naa.PitchAug(sampling_rate=sampling_rate, factor=(0.2, 1.5)),
naa.ShiftAug(sampling_rate=sampling_rate, duration=2),
naa.SpeedAug(factor=(1.5, 3))
])
augmented_audio = flow.augment(audio)
self.assertFalse(np.array_equal(audio, augmented_audio))
self.assertTrue(len(audio), len(augmented_audio))
def test_n_output_without_augmentation(self):
texts = ['AAAAAAAAAAA AAAAAAAAAAAAAA']
flows = [
naf.Sequential([nac.OcrAug(), nac.OcrAug()]),
naf.Sometimes(
[nac.RandomCharAug(), nac.RandomCharAug()], pipeline_p=0.00001)
]
for flow in flows:
for text in texts:
at_least_one_equal = False
for _ in range(5):
augmented_texts = flow.augment(text, n=3)
if len(augmented_texts
) == 1 and augmented_texts[0] == text:
at_least_one_equal = True
break
self.assertTrue(at_least_one_equal)
self.assertLess(0, len(flows))
self.assertLess(0, len(texts))
def test_spectrogram(self):
# https://freewavesamples.com/yamaha-v50-rock-beat-120-bpm
sample_wav_file = os.path.abspath(
os.path.join(os.path.dirname(__file__), '..', '..', 'data',
'Yamaha-V50-Rock-Beat-120bpm.wav'))
mel_spectrogram = LoadUtil.load_mel_spectrogram(sample_wav_file,
n_mels=128)
flow = naf.Sequential([
nas.FrequencyMaskingAug(mask_factor=50),
nas.TimeMaskingAug(mask_factor=20),
nas.TimeMaskingAug(mask_factor=30)
])
augmented_mel_spectrogram = flow.augment(mel_spectrogram)
for aug in flow:
if aug.name == 'FrequencyMasking_Aug':
self.assertEqual(
len(mel_spectrogram[aug.model.f0]),
np.count_nonzero(mel_spectrogram[aug.model.f0]))
self.assertEqual(
0,
np.count_nonzero(augmented_mel_spectrogram[aug.model.f0]))
elif aug.name == 'TimeMasking_Aug':
self.assertEqual(
len(mel_spectrogram[:, aug.model.t0]),
np.count_nonzero(mel_spectrogram[:, aug.model.t0]))
self.assertEqual(
0,
np.count_nonzero(augmented_mel_spectrogram[:,
aug.model.t0]))
else:
# Unexpected flow
self.assertFalse(True)
self.assertTrue(len(flow) > 0)
def augment_n(data, N=1):
pbar = tqdm(desc='Augmenting Data N={}'.format(N),
total=data.shape[0],
leave=False)
# random synonym replacement
# aug = naw.SynonymAug(aug_max=4, stopwords=stop_words())
aug = naf.Sequential([
# naw.ContextualWordEmbsAug(
# 'bert-base-uncased',
# aug_max=5,
# stopwords=stop_words(),
# device='cuda',
# optimize=True
# ),
naw.ContextualWordEmbsAug('bert-base-uncased',
aug_max=3,
stopwords=stop_words(),
device='cuda',
optimize=True,
action='insert'),
naw.SynonymAug(aug_max=4, stopwords=stop_words())
])
results = []
for row in data:
t, s = augment(row[1], row[2], aug, N)
augs = []
for j, t in enumerate(t):
augs.append([row[0] + str(j), t, s[j], row[3]])
if len(augs) > 0:
results.append(np.array(augs))
pbar.update()
results.append(data)
pbar.clear()
pbar.close()
return np.concatenate(results, axis=0)
def test_n_output_without_augmentation(self):
texts = ['AAAAAAAAAAA AAAAAAAAAAAAAA']
flows = [
naf.Sequential([nac.OcrAug(), nac.OcrAug()]),
naf.Sometimes(
[nac.RandomCharAug(), nac.RandomCharAug()], pipeline_p=0.00001)
]
for flow in flows:
for text in texts:
for _ in range(5):
augmented_texts = flow.augment(text, n=3)
all_not_equal = False
for augmented_text in augmented_texts:
if augmented_text != text:
all_not_equal = True
break
if all_not_equal:
break
self.assertFalse(all_not_equal)
self.assertLess(0, len(flows))
self.assertLess(0, len(texts))
def test_dry_run(self):
flow = naf.Sequential()
results = flow.augment([])
self.assertEqual(0, len(results))
def build_augments(self):
self.augs = naf.Sequential([
naw.SynonymAug(aug_src='wordnet'),
# naw.ContextualWordEmbsAug(model_path='bert-base-uncased', action="insert"),
# naw.RandomWordAug(action="swap")
])
def __call__(self, data):
transforms = naf.Sequential(
[self.get_random_freq_mask(),
self.get_random_time_mask()])
return transforms.augment(data)
try:
os.makedirs(path)
except OSError:
print("Creation of the directory %s failed" % path)
else:
print("Creation of the directory %s success" % path)
if os.path.exists(csv_file_path):
input_directory = Path(csv_file_path)
for my_filename in input_directory.glob("*_AUDIO_*.wav"):
audio, sampling_rate = librosa.load(my_filename)
VisualWave.visual('Original', audio, sampling_rate)
flow = naf.Sequential([
naa.NoiseAug(),
naa.PitchAug(sampling_rate=sampling_rate,
pitch_factor=1.5),
naa.ShiftAug(sampling_rate=sampling_rate, shift_max=2),
naa.SpeedAug(speed_factor=1.5)
])
augmented_audio = flow.augment(audio)
VisualWave.visual('augment', augmented_audio,
sampling_rate)
my_filename = my_filename.stem
librosa.output.write_wav(path + my_filename + '.wav',
augmented_audio,
sampling_rate,
norm=False)
except ValueError:
print("Skipping the following line: ", row[0])
csvFile.close()