def test_without_noise(self): aug = naa.MaskAug(sampling_rate=self.sampling_rate, mask_with_noise=False) augmented_audio = aug.augment(self.audio) self.assertFalse(np.array_equal(self.audio, augmented_audio)) self.assertEqual(len(self.audio), len(augmented_audio))
def test_with_noise(self): audio, sampling_rate = librosa.load(self.sample_wav_file) aug = naa.MaskAug(sampling_rate=sampling_rate, mask_with_noise=True) augmented_audio = aug.augment(audio) self.assertFalse(np.array_equal(audio, augmented_audio)) self.assertEqual(len(audio), len(augmented_audio))
def test_zone(self): zone = (0, 1) coverage = 1. aug = naa.MaskAug(sampling_rate=self.sampling_rate, zone=zone, coverage=coverage, mask_with_noise=False) aug.model.stateless = False augmented_audio = aug.augment(self.audio) reconstruct_augmented_audio = np.concatenate( (self.audio[:aug.model.start_pos], aug.model.aug_data, self.audio[aug.model.end_pos:]), axis=0) self.assertTrue(np.array_equal(augmented_audio, reconstruct_augmented_audio)) self.assertTrue(len(aug.model.aug_data), int(len(self.audio) * (zone[1] - zone[0]) * coverage))
def test_zone(self): zone = (0, 1) coverage = 1. decimal = 8 aug = naa.MaskAug(sampling_rate=self.sampling_rate, zone=zone, coverage=coverage) aug.model.stateless = False augmented_audio = aug.augment(self.audio) reconstruct_augmented_audio = np.concatenate( (self.audio[:aug.model.start_pos], aug.model.aug_data, self.audio[aug.model.end_pos:]) , axis=0).astype(np.float32) augmented_audio = np.round(augmented_audio, decimals=decimal) reconstruct_augmented_audio = np.round(reconstruct_augmented_audio, decimals=decimal) self.assertTrue(np.array_equal(augmented_audio, reconstruct_augmented_audio)) self.assertTrue(len(aug.model.aug_data), int(len(self.audio) * (zone[1] - zone[0]) * coverage))
def test_coverage_and_zone(self): params = [ ((0.3, 0.7), 1), ((0, 1), 1) ] for zone, coverage in params: augs = [ naa.LoudnessAug(zone=zone, coverage=coverage, stateless=False), naa.MaskAug(zone=zone, coverage=coverage, stateless=False), naa.NoiseAug(zone=zone, coverage=coverage, stateless=False), naa.PitchAug(zone=zone, coverage=coverage, stateless=False, sampling_rate=self.sampling_rate), naa.SpeedAug(zone=zone, coverage=coverage, stateless=False), naa.VtlpAug(zone=zone, coverage=coverage, stateless=False, sampling_rate=self.sampling_rate), naa.NormalizeAug(zone=zone, coverage=coverage, stateless=False), naa.PolarityInverseAug(zone=zone, coverage=coverage, stateless=False) ] for aug in augs: aug_data = aug.augment(self.audio) self.assertTrue(len(aug_data[aug.start_pos:aug.end_pos]), int(len(self.audio) * (zone[1] - zone[0]) * coverage))
def mask(data, sampling_rate): aug = naa.MaskAug(sampling_rate=sampling_rate, mask_with_noise=False) return aug.augment(data)
def test_empty_input(self): audio = np.array([]) aug = naa.MaskAug(sampling_rate=44100) augmented_audio = aug.augment(audio) self.assertTrue(np.array_equal(audio, augmented_audio))
def add_mask(data, sr): aug = naa.MaskAug(sampling_rate=sr, mask_with_noise=False) return aug.augment(data)