def __init__(self, opt):
"""Initialize this dataset class.
Parameters:
opt (Option class) -- stores all the experiment flags; needs to be a subclass of BaseOptions
"""
# save the option and dataset root
BaseDataset.__init__(self, opt)
# get the audio path
# TODO: search genre and maintain a list
FMA
# self.A_dir = os.path.isdir(os.path.join(self.root, opt.A_subdir))
# self.B_dir = os.path.isdir(os.path.join(self.root, opt.B_subdir))
self.A_paths = sorted(make_dataset(self.A_dir, opt.max_dataset_size))
self.B_paths = sorted(make_dataset(self.B_dir, opt.max_dataset_size))
self.A_size = len(self.A_paths)
self.B_size = len(self.B_paths)
self.sample_rate = opt.sample_rate
self.nfft = opt.nfft
self.mel = opt.mel
def initialize(self, opt):
self.opt = opt
self.DirClean = opt.PathClean
self.DirNoise = opt.PathNoise
self.snr = opt.snr
self.Clean = make_dataset(self.DirClean, opt)
self.Noise = make_dataset(self.DirNoise, opt)
self.SR = opt.SR
self.hop = opt.hop
self.nfft = self.opt.nfft
def initialize(self, opt):
self.opt = opt
self.Dir = opt.Path
self.Data, self.Labels, self.Fnames = make_dataset(opt)
self.pinkNoise = getMel(
'/home/diggerdu/dataset/tfsrc/train/_background_noise_/pink_noise.wav',
opt)
self.oriLen = len(self.Data)
self.SR = opt.SR
self.hop = opt.hop
self.nfft = self.opt.nfft
self.table = [
'yes', 'no', 'up', 'down', 'left', 'right', 'on', 'off', 'stop',
'go', 'silence', 'unknown'
]
if opt.isTrain:
self.augFuncList = [lambda x: x, freqFilter()]
else:
self.augFuncList = [lambda x: x]
def __init__(self, opt):
"""Initialize this dataset class.
Parameters:
opt (Option class) -- stores all the experiment flags; needs to be a subclass of BaseOptions
A few things can be done here.
- save the options (have been done in BaseDataset)
- get image paths and meta information of the dataset.
- define the image transformation.
"""
# save the option and dataset root
BaseDataset.__init__(self, opt)
# get the audio path
self.dirpath = os.path.isdir('{}/{}', self.root, opt.subdir)
# TODO: Use cache if it exists
# self.cachedir = os.path.join(self.dirpath, '__cache')
# mkdir(self.cachedir)
# metadata_file = os.path.join(self.cachedir, 'metadata.csv')
# if os.path.exist(metadata_file):
# with open(metadata_file, 'r') as f:
# metadata = json.load(f)
# else:
# metadata = dict()
self.audio_paths = sorted(
make_dataset(self.dirpath, opt.max_dataset_size))
self.frame_paths = list()
self.sample_rate = opt.sample_rate
self.length = opt.length
self.stride = opt.stride
self.nfft = opt.nfft
self.subdir = opt.subdir
self.mel = opt.mel
for p in self.audio_paths:
# Load audio
filename = os.path.splitext(os.path.basename(p))[0]
y, sr = sf.read(p, dtype='float32')
t = librosa.get_duration(y=y, sr=sr)
# Resample Audio
if sr != self.sample_rate:
y = librosa.resample(y, sr, self.sample_rate)
# Pad the audio
l = t % opt.stride
if l != 0:
t = t + 15 - l
y = librosa.util.fix_length(y, t * self.sample_rate)
# Length Check
if t < self.length:
print('Skipping {} due to short content length'.format(p))
continue
# Compute frames and store
frames = self.frame(y, sr, length=self.length, stride=self.stride)
# cannot store all the frames; store it into files
for i in len(frames):
fp = os.path.join(cachedir, '{}.{}.npy'.format(filename, i))
np.save(fp, frames[:, i])
self.frame_paths.append({
'file': filename,
'frame': i,
'path': fp
})