def test_n_output_audio(self):
audio, sampling_rate = AudioLoader.load_audio(self.sample_wav_file)
flows = [
naf.Sequential(
[naa.CropAug(sampling_rate=sampling_rate),
naa.LoudnessAug()]),
naf.Sometimes(
[naa.CropAug(sampling_rate=sampling_rate),
naa.LoudnessAug()],
pipeline_p=0.9),
naf.Sequential([
naf.Sequential([
naa.CropAug(sampling_rate=sampling_rate),
naa.LoudnessAug()
]),
naf.Sometimes([
naa.CropAug(sampling_rate=sampling_rate),
naa.LoudnessAug()
],
pipeline_p=0.9)
])
]
for flow in flows:
augmented_audios = flow.augment(audio, n=3)
self.assertGreater(len(augmented_audios), 1)
for augmented_audio in augmented_audios:
self.assertFalse(np.array_equal(audio, augmented_audio))
self.assertLess(0, len(flows))
def test_subsitute(self):
audio, sampling_rate = librosa.load(self.sample_wav_file)
aug = naa.CropAug(sampling_rate=sampling_rate)
augmented_audio = aug.augment(audio)
self.assertNotEqual(len(audio), len(augmented_audio))
def test_coverage(self):
aug = naa.CropAug(sampling_rate=self.sampling_rate, coverage=0.1)
augmented_data = aug.augment(self.audio)
audio_size = len(self.audio)
augmented_size = len(augmented_data)
expected_crop_size = len(
self.audio) * (aug.model.zone[1] - aug.model.zone[0]) * 0.1
self.assertTrue(
-1 <= audio_size - augmented_size - expected_crop_size <= 1)
def test_multi_thread(self):
n = 3
augs = [
naa.CropAug(sampling_rate=self.sampling_rate),
naa.PitchAug(sampling_rate=self.sampling_rate)
]
for num_thread in [1, 3]:
for aug in augs:
augmented_data = aug.augment(self.audio, n=n, num_thread=num_thread)
self.assertEqual(len(augmented_data), n)
def test_multi_thread(self):
audio, sampling_rate = AudioLoader.load_audio(self.sample_wav_file)
n = 3
augs = [
naa.CropAug(sampling_rate=sampling_rate),
naa.PitchAug(sampling_rate=sampling_rate)
]
for num_thread in [1, 3]:
for aug in augs:
augmented_data = aug.augment(audio, n=n, num_thread=num_thread)
self.assertEqual(len(augmented_data), n)
def setUpClass(cls):
env_config_path = os.path.abspath(os.path.join(
os.path.dirname(__file__), '..', '..', '.env'))
load_dotenv(env_config_path)
# https://freewavesamples.com/yamaha-v50-rock-beat-120-bpm
cls.sample_wav_file = os.path.join(
os.environ.get("TEST_DIR"), 'res', 'audio', 'Yamaha-V50-Rock-Beat-120bpm.wav'
)
cls.audio, cls.sampling_rate = AudioLoader.load_audio(cls.sample_wav_file)
cls.audio_augs = [
naa.CropAug(sampling_rate=cls.sampling_rate),
naa.SpeedAug(),
]
def test_duration(self):
duration = 1
audio_size = len(self.audio)
for _ in range(10):
aug = naa.CropAug(sampling_rate=self.sampling_rate,
duration=duration,
stateless=False)
aug_data = aug.augment(self.audio)
aug_size = len(aug_data)
expected_crop_size = self.sampling_rate * duration
self.assertGreater(audio_size, aug_size)
self.assertEqual(len(self.audio[aug.start_pos:aug.end_pos]),
expected_crop_size)
def process_audio_files_with_aug(filename, dirpath):
sr = 16000
audio_array, sample_rate = librosa.load(os.path.join(dirpath, 'flac', filename), sr=sr)
aug_crop = naa.CropAug(sampling_rate=sr)
audio_array_crop = aug_crop.augment(audio_array)
aug_loud = naa.LoudnessAug(loudness_factor=(2, 5))
audio_array_loud = aug_loud.augment(audio_array)
aug_noise = naa.NoiseAug(noise_factor=0.03)
audio_array_noise = aug_noise.augment(audio_array)
audio_array_list= [audio_array,audio_array_crop,audio_array_loud,
audio_array_noise ]
out_list = convert_audio_to_processed_list(audio_array_list, filename, dirpath)
return out_list
def setUpClass(cls):
env_config_path = os.path.abspath(
os.path.join(os.path.dirname(__file__), '..', '..', '.env'))
load_dotenv(env_config_path)
# https://freewavesamples.com/yamaha-v50-rock-beat-120-bpm
cls.sample_wav_file = os.environ.get(
"DATA_DIR") + 'Yamaha-V50-Rock-Beat-120bpm.wav'
cls.audio, cls.sampling_rate = librosa.load(cls.sample_wav_file)
cls.textual_augs = [
nac.RandomCharAug(),
naw.ContextualWordEmbsAug(),
nas.ContextualWordEmbsForSentenceAug()
]
cls.audio_augs = [
naa.CropAug(sampling_rate=cls.sampling_rate),
naa.SpeedAug(),
]
def process_audio_files_with_aug(self, audio_path):
sr = 16000
audio_array, sample_rate = librosa.load(audio_path, sr=sr)
aug_crop = naa.CropAug(sampling_rate=sr)
audio_array_crop = aug_crop.augment(audio_array)
aug_loud = naa.LoudnessAug(loudness_factor=(2, 5))
audio_array_loud = aug_loud.augment(audio_array)
aug_noise = naa.NoiseAug(noise_factor=0.03)
audio_array_noise = aug_noise.augment(audio_array)
mel_spec_array_load = melspectrogram(audio_array_loud, hparams=hparams)
mel_spec_array_noise = melspectrogram(audio_array_noise,
hparams=hparams)
audio_array_list = [mel_spec_array_load, mel_spec_array_noise]
# audio_array_list= [audio_array_crop,audio_array_loud,
# audio_array_noise ]
return audio_array_list
def test_substitute(self):
aug = naa.CropAug(sampling_rate=self.sampling_rate)
augmented_audio = aug.augment(self.audio)
self.assertNotEqual(len(self.audio), len(augmented_audio))
def test_empty_input(self):
audio = np.array([])
aug = naa.CropAug(sampling_rate=self.sampling_rate)
augmented_audio = aug.augment(audio)
self.assertTrue(np.array_equal(audio, augmented_audio))
def crop(data, sampling_rate):
aug = naa.CropAug(sampling_rate=sampling_rate)
return aug.augment(data)
def crop_audio(data, sr):
aug = naa.CropAug(sampling_rate=sr)
return aug.augment(data)
