rate=SAMPLE_RATE,
data=augmented_samples)
# AddGaussianNoise
augmenter = Compose(
[AddGaussianNoise(min_amplitude=0.001, max_amplitude=0.015, p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "AddGaussianNoise_{:03d}.wav".format(i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# TimeStretch
augmenter = Compose([TimeStretch(min_rate=0.8, max_rate=1.25, p=1.0)])
for i in range(5):
output_file_path = os.path.join(output_dir,
"TimeStretch_{:03d}.wav".format(i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# PitchShift
augmenter = Compose([PitchShift(min_semitones=-4, max_semitones=4, p=1.0)])
for i in range(5):
output_file_path = os.path.join(output_dir,
"PitchShift_{:03d}.wav".format(i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
"num_runs": 5
},
{
"instance":
Shift(min_fraction=-0.5, max_fraction=0.5, rollover=False, p=1.0),
"num_runs":
5,
"name":
"ShiftWithoutRollover",
},
{
"instance": TimeMask(p=1.0),
"num_runs": 5
},
{
"instance": TimeStretch(min_rate=0.8, max_rate=1.25, p=1.0),
"num_runs": 5
},
{
"instance": Trim(p=1.0),
"num_runs": 1
},
]
for sound_file_path in sound_file_paths:
samples, sample_rate = load_sound_file(sound_file_path,
sample_rate=None,
mono=False)
if len(samples.shape) == 2 and samples.shape[0] > samples.shape[1]:
samples = samples.transpose()
from python_speech_features import mfcc
import scipy.io.wavfile as wav
import matplotlib.pyplot as plt
from matplotlib import cm
import numpy as np
import os
import random
from tqdm import tqdm
from audiomentations import Compose, AddGaussianNoise, TimeStretch, PitchShift, Shift
augmenter = Compose([
AddGaussianNoise(min_amplitude=0.001, max_amplitude=0.015, p=0.5),
TimeStretch(min_rate=0.8, max_rate=1.25, p=0.5),
PitchShift(min_semitones=-4, max_semitones=4, p=0.5),
Shift(min_fraction=-0.5, max_fraction=0.5, p=0.5),
])
def load_noise(path='/home/CAIL/Speaker_R/data/voice/background_noise/'):
noise = []
files = os.listdir(path)
for f in files:
filename = f
if ('wav' not in filename):
continue
f = os.path.join(path, f)
(rate, sig) = wav.read(f)
noise.append(sig)
return noise
def applyTransformations(fileName, output_dir, auxiliarSoundsDir):
name = fileName.split(".")[0].split("/")[-1]
samples = load_wav_file(fileName)
# AddImpulseResponse
augmenter = Compose([
AddImpulseResponse(p=1.0,
ir_path=os.path.join(auxiliarSoundsDir,
"helperSounds/ir"))
])
output_file_path = os.path.join(
output_dir, "{}_AddImpulseResponse_{:03d}.wav".format(name, 0))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path, rate=SAMPLE_RATE, data=augmented_samples)
# FrequencyMask
augmenter = Compose([FrequencyMask(p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_FrequencyMask_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# TimeMask
augmenter = Compose([TimeMask(p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_TimeMask_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# AddGaussianSNR
augmenter = Compose([AddGaussianSNR(p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_AddGaussianSNR_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# AddGaussianNoise
augmenter = Compose(
[AddGaussianNoise(min_amplitude=0.001, max_amplitude=0.015, p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_AddGaussianNoise_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# TimeStretch
augmenter = Compose([TimeStretch(min_rate=0.8, max_rate=1.25, p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_TimeStretch_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# PitchShift
augmenter = Compose([PitchShift(min_semitones=-4, max_semitones=4, p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_itchShift_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# Shift
augmenter = Compose([Shift(min_fraction=-0.5, max_fraction=0.5, p=1.0)])
for i in range(5):
output_file_path = os.path.join(output_dir,
"{}_Shift_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# Shift without rollover
augmenter = Compose(
[Shift(min_fraction=-0.5, max_fraction=0.5, rollover=False, p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_ShiftWithoutRollover_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# Normalize
augmenter = Compose([Normalize(p=1.0)])
output_file_path = os.path.join(output_dir,
"{}_Normalize_{:03d}.wav".format(name, 0))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path, rate=SAMPLE_RATE, data=augmented_samples)
# ClippingDistortion
augmenter = Compose([ClippingDistortion(p=1.0)])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_ClippingDistortion_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# AddBackgroundNoise
augmenter = Compose([
AddBackgroundNoise(sounds_path=os.path.join(
auxiliarSoundsDir, "helperSounds/background_noises"),
p=1.0)
])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_AddBackgroundNoise_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
# AddShortNoises
augmenter = Compose([
AddShortNoises(
sounds_path=os.path.join(auxiliarSoundsDir,
"helperSounds/short_noises"),
min_snr_in_db=0,
max_snr_in_db=8,
min_time_between_sounds=2.0,
max_time_between_sounds=4.0,
burst_probability=0.4,
min_pause_factor_during_burst=0.01,
max_pause_factor_during_burst=0.95,
min_fade_in_time=0.005,
max_fade_in_time=0.08,
min_fade_out_time=0.01,
max_fade_out_time=0.1,
p=1.0,
)
])
for i in range(5):
output_file_path = os.path.join(
output_dir, "{}_AddShortNoises_{:03d}.wav".format(name, i))
augmented_samples = augmenter(samples=samples, sample_rate=SAMPLE_RATE)
wavfile.write(output_file_path,
rate=SAMPLE_RATE,
data=augmented_samples)
def __getitem__(self, idx):
augment = Compose([
TimeStretch(min_rate=0.8, max_rate=1.25, p=0.5),
Shift(min_fraction=-0.5, max_fraction=0.5, p=0.5, rollover=False)
])
temp1 = random.randint(0, 1)
temp2 = random.randint(0, 1)
temp3 = random.randint(0, 1)
self.anchor = str(self.data.iloc[idx, 0])
self.positive = self.data.iloc[idx, 1]
self.negative = self.data.iloc[idx, 2]
self.signalAnchor, self.srAnchor = torchaudio.load(self.anchor)
self.signalPositive, self.srPositive = torchaudio.load(self.positive)
self.signalNegative, self.srNegative = torchaudio.load(self.negative)
if (temp1 == 1):
self.signalAnchor = torch.from_numpy(
augment(samples=self.signalAnchor.numpy(),
sample_rate=self.srAnchor))
if (temp2 == 1):
self.signalPositive = torch.from_numpy(
augment(samples=self.signalPositive.numpy(),
sample_rate=self.srPositive))
if (temp3 == 1):
self.signalNegative = torch.from_numpy(
augment(samples=self.signalNegative.numpy(),
sample_rate=self.srNegative))
self.spectogramAnchor = torchaudio.transforms.Spectrogram(
n_fft=320, hop_length=160, win_length=320)(self.signalAnchor)
self.logSpectogramAnchor = torchaudio.transforms.AmplitudeToDB()(
self.spectogramAnchor)
self.spectogramPositive = torchaudio.transforms.Spectrogram(
n_fft=320, hop_length=160, win_length=320)(self.signalPositive)
self.logSpectogramPositive = torchaudio.transforms.AmplitudeToDB()(
self.spectogramPositive)
self.spectogramNegative = torchaudio.transforms.Spectrogram(
n_fft=320, hop_length=160, win_length=320)(self.signalNegative)
self.logSpectogramNegative = torchaudio.transforms.AmplitudeToDB()(
self.spectogramNegative)
#self.tempImgAnchor=torchvision.transforms.ToPILImage()(self.logSpectogramAnchor)
#self.tempImgAnchor=self.tempImgAnchor.convert("RGB")
#self.spectogramAnchorImageTensor=self.vision_transform(self.tempImgAnchor)
#self.tempImgPositive=torchvision.transforms.ToPILImage()(self.logSpectogramPositive)
#self.tempImgPositive=self.tempImgPositive.convert("RGB")
#self.spectogramPositiveImageTensor=self.vision_transform(self.tempImgPositive)
#self.tempImgNegative=torchvision.transforms.ToPILImage()(self.logSpectogramNegative)
#self.tempImgNegative=self.tempImgNegative.convert("RGB")
#self.spectogramNegativeImageTensor=self.vision_transform(self.tempImgNegative)
return self.logSpectogramAnchor, self.logSpectogramPositive, self.logSpectogramNegative