def augment_audio(data, sampling_rate, loudness_factor, speed_range, pitch_range, shift_max, noise_factor):
"""
OPTIONS:
LOUDNESS AUGMENTER: naa.LoudnessAug()
MASK AUGMENTER: naa.MaskAug()
SPEED AUGMENTER: naa.SpeedAug()
SHIFT AUGMENTER: naa.ShiftAug()
NOISE AUGMENTER: naa.NoiseAug()
CROP AUGMENTER: naa.CropAug()
PITCH AUGMENTER_ naa.PitchAug()
"""
flow = naf.Sequential([naa.LoudnessAug(loudness_factor),
naa.SpeedAug(speed_range),
naa.PitchAug(sampling_rate = sampling_rate, pitch_range = pitch_range),
naa.ShiftAug(sampling_rate = sampling_rate, shift_max = shift_max),
naa.NoiseAug(noise_factor),
])
augmented_audio = flow.augment(data)
return augmented_audio
def testSubsitute(self):
audio, sampling_rate = librosa.load(self.sample_wav_file)
aug = naa.SpeedAug(speed_factor=1.5)
augmented_audio = aug.augment(audio)
self.assertFalse(np.array_equal(audio, augmented_audio))
self.assertTrue(len(audio), len(augmented_audio))
def test_substitute(self):
for _ in range(10):
aug = naa.SpeedAug(stateless=False)
aug_data = aug.augment(self.audio)
if aug.aug_factor < 1:
self.assertGreater(len(aug_data), len(self.audio))
else:
self.assertLess(len(aug_data), len(self.audio))
def test_audio(self):
audio, sampling_rate = AudioLoader.load_audio(self.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 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 gen_augmented(original, sample_rate):
# list of augmentors available from the nlpaug library
augment_list = [
#naa.CropAug(sampling_rate=sample_rate)
naa.NoiseAug(),
naa.SpeedAug(),
naa.LoudnessAug(factor=(0.5, 2)),
naa.VtlpAug(sampling_rate=sample_rate, zone=(0.0, 1.0)),
naa.PitchAug(sampling_rate=sample_rate, factor=(-1, 3))
]
# sample augmentation randomly
aug_idx = random.randint(0, len(augment_list) - 1)
augmented_data = augment_list[aug_idx].augment(original)
return augmented_data
def test_zone(self):
zone = (0, 1)
coverage = 1.
for _ in range(10):
aug = naa.SpeedAug(zone=zone, coverage=coverage)
aug.model.stateless = False
aug.augment(self.audio)
if aug.model.aug_factor < 1:
self.assertGreater(
len(aug.model.aug_data),
len(self.audio[aug.model.start_pos:aug.model.end_pos]))
else:
self.assertLess(
len(aug.model.aug_data),
len(self.audio[aug.model.start_pos:aug.model.end_pos]))
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 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 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 aug_speed(data, speed_range=(0.9, 1.1)): aug = naa.SpeedAug(speed_range=speed_range) augmented_data = aug.augment(data) return augmented_data
def speed(data):
aug = naa.SpeedAug()
return aug.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()
def add_speed(data, sr):
aug = naa.SpeedAug()
return aug.augment(data)
