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.VtlpAug(sampling_rate=self.sampling_rate,
zone=zone,
coverage=coverage)
aug.model.stateless = False
aug.augment(self.audio)
self.assertGreater(
len(self.audio[aug.model.start_pos:aug.model.end_pos]),
len(aug.model.aug_data))
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 test_substitute(self):
for _ in range(10):
aug = naa.VtlpAug(sampling_rate=self.sampling_rate,
stateless=False)
augmented_audio = aug.augment(self.audio)
self.assertGreater(len(self.audio), len(augmented_audio))
import nlpaug.augmenter.audio as naa
import librosa
import glob
from tqdm import tqdm
import os
wlist=glob.glob(r'../IEMOCAP/*.wav')
targetDir='../IEMOCAP/'
aug = naa.VtlpAug(16000, zone=(0.0, 1.0), coverage=1, duration=None, fhi=4800, factor=(0.8, 1.2))
for w in tqdm(wlist):
for i in range(7):
wav,_=librosa.load(w,16000)
wavAug=aug.augment(wav)
wavName=os.path.basename(w)
librosa.output.write_wav(targetDir+wavName+'.'+str(i+1),wavAug,16000)
def vtlp(data, sampling_rate):
aug = naa.VtlpAug(sampling_rate=sampling_rate)
return aug.augment(data)
def add_vltk(data, sr):
aug = naa.VtlpAug(sampling_rate=sr)
return aug.augment(data)