def get_example(self, i):
train = chainer.config.train
inputs = self.inputs[i]
p_input, p_target, p_indexes = inputs.in_feature_path, inputs.out_feature_path, inputs.indexes_path
indexes = AlignIndexes.load(p_indexes)
# input feature
f_in = AcousticFeature.load(p_input)
f_in = f_in.indexing(indexes.indexes1)
input = encode_feature(f_in, targets=self.config.in_features)
# target feature
f_tar = AcousticFeature.load(p_target)
f_tar = f_tar.indexing(indexes.indexes2)
target = encode_feature(f_tar, targets=self.config.out_features)
mask = encode_feature(make_mask(f_tar),
targets=self.config.out_features)
# padding
seed = numpy.random.randint(2**31)
input = random_pad(input,
seed=seed,
min_size=self.config.train_crop_size)
target = random_pad(target,
seed=seed,
min_size=self.config.train_crop_size)
mask = random_pad(mask,
seed=seed,
min_size=self.config.train_crop_size)
# crop
seed = numpy.random.randint(2**31)
input = random_crop(input,
seed=seed,
crop_size=self.config.train_crop_size)
target = random_crop(target,
seed=seed,
crop_size=self.config.train_crop_size)
mask = random_crop(mask,
seed=seed,
crop_size=self.config.train_crop_size)
if train:
input = add_noise(input,
p_global=self.config.input_global_noise,
p_local=self.config.input_local_noise)
target = add_noise(target,
p_global=self.config.target_global_noise,
p_local=self.config.target_local_noise)
return dict(
input=input,
target=target,
mask=mask,
)
def get_example(self, i):
train = chainer.config.train
p_x = self.x_paths[numpy.random.randint(len(self.x_paths))]
p_y = self.y_paths[numpy.random.randint(len(self.y_paths))]
f_x = AcousticFeature.load(p_x)
x = encode_feature(f_x, targets=self.config.in_features)
f_y = AcousticFeature.load(p_y)
y = encode_feature(f_y, targets=self.config.out_features)
mask_x = encode_feature(make_mask(f_x),
targets=self.config.in_features)
mask_y = encode_feature(make_mask(f_y),
targets=self.config.out_features)
# padding
seed = numpy.random.randint(2**31)
x = random_pad(x, seed=seed, min_size=self.config.train_crop_size)
mask_x = random_pad(mask_x,
seed=seed,
min_size=self.config.train_crop_size)
seed = numpy.random.randint(2**31)
y = random_pad(y, seed=seed, min_size=self.config.train_crop_size)
mask_y = random_pad(mask_y,
seed=seed,
min_size=self.config.train_crop_size)
# crop
seed = numpy.random.randint(2**31)
x = random_crop(x, seed=seed, crop_size=self.config.train_crop_size)
mask_x = random_crop(mask_x,
seed=seed,
crop_size=self.config.train_crop_size)
seed = numpy.random.randint(2**31)
y = random_crop(y, seed=seed, crop_size=self.config.train_crop_size)
mask_y = random_crop(mask_y,
seed=seed,
crop_size=self.config.train_crop_size)
if train:
x = add_noise(x,
p_global=self.config.input_global_noise,
p_local=self.config.input_local_noise)
y = add_noise(y,
p_global=self.config.target_global_noise,
p_local=self.config.target_local_noise)
return dict(
x=x,
y=y,
mask_x=mask_x,
mask_y=mask_y,
)
def generate_align_indexes(pair_path: Tuple[Path, Path]):
path1, path2 = pair_path
if path1.stem != path2.stem:
print('warning: the file names are different', path1, path2)
out = Path(arguments.output, path1.stem + '.npy')
if out.exists() and not arguments.enable_overwrite:
return
feature1 = AcousticFeature.load(path=path1)
feature2 = AcousticFeature.load(path=path2)
align_indexes = AlignIndexes.extract(feature1, feature2, dtype=arguments.dtype)
# save
align_indexes.save(path=out, validate=True, ignores=arguments.ignore_feature)
def generate_aligned_wave(
pair_path: Tuple[Path, Path, Path],
sampling_rate: int,
frame_period: float,
alpha: float,
):
path_feature1, path_feature2, path_indexes = pair_path
if path_feature1.stem != path_feature2.stem:
print('warning: the file names are different', path_feature1,
path_feature2)
if path_feature1.stem != path_indexes.stem:
print('warning: the file names are different', path_feature1,
path_indexes)
out = Path(arguments.output, path_indexes.stem + '.wav')
if arguments.disable_overwrite:
return
feature1 = AcousticFeature.load(path=path_feature1)
feature2 = AcousticFeature.load(path=path_feature2)
feature1.sp = AcousticFeature.mc2sp(feature1.mc,
sampling_rate=sampling_rate,
alpha=alpha)
feature2.sp = AcousticFeature.mc2sp(feature2.mc,
sampling_rate=sampling_rate,
alpha=alpha)
feature1.ap = AcousticFeature.decode_ap(feature1.coded_ap,
sampling_rate=sampling_rate)
feature2.ap = AcousticFeature.decode_ap(feature2.coded_ap,
sampling_rate=sampling_rate)
align_indexes = AlignIndexes.load(path=path_indexes)
align_indexes.feature1 = feature1
align_indexes.feature2 = feature2
wave1 = align_indexes.get_aligned_feature1().decode(
sampling_rate=sampling_rate, frame_period=frame_period)
wave2 = align_indexes.get_aligned_feature2().decode(
sampling_rate=sampling_rate, frame_period=frame_period)
# save
y = numpy.vstack([wave1.wave, wave2.wave])
librosa.output.write_wav(str(out), y, sr=sampling_rate)
def load_f0(path: Path):
feature = AcousticFeature.load(path=path)
return feature.f0
def load_acoustic_feature(self, path: Path):
return AcousticFeature.load(path)