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 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_multi_thread(self):
mel_spectrogram = LoadUtil.load_mel_spectrogram(self.sample_wav_file, n_mels=128)
n = 3
augs = [
nas.FrequencyMaskingAug(mask_factor=80),
nas.TimeMaskingAug(mask_factor=80)
]
for num_thread in [1, 3]:
for aug in augs:
augmented_data = aug.augment(mel_spectrogram, n=n, num_thread=num_thread)
self.assertEqual(len(augmented_data), n)
def test_substitute(self):
freq_mask_para = 80
mel_spectrogram = LoadUtil.load_mel_spectrogram(self.sample_wav_file,
n_mels=128)
aug = FrequencyMaskingAug(mask_factor=freq_mask_para)
augmented_mel_spectrogram = aug.augment(mel_spectrogram)
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]))
def test_substitute(self):
time_mask_para = 80
mel_spectrogram = LoadUtil.load_mel_spectrogram(
self.sample_wav_file, n_mels=self.num_of_freq_channel)
aug = TimeMaskingAug(mask_factor=time_mask_para)
augmented_mel_spectrogram = aug.augment(mel_spectrogram)
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]))
