def __init__(self, in_features, kernel_size, padding, **kwargs):
super(ResBlock2d, self).__init__(**kwargs)
self.conv1 = dygraph.Conv2D(num_channels=in_features,
num_filters=in_features,
filter_size=kernel_size,
padding=padding)
self.conv2 = dygraph.Conv2D(num_channels=in_features,
num_filters=in_features,
filter_size=kernel_size,
padding=padding)
self.norm1 = dygraph.BatchNorm(num_channels=in_features, momentum=0.1)
self.norm2 = dygraph.BatchNorm(num_channels=in_features, momentum=0.1)
def get_norm(norm_type, channels_num):
if norm_type == "AN":
return AffineNormalization(channels_num)
elif norm_type == "BN":
return dg.BatchNorm(channels_num)
else:
raise NotImplementedError()
def __init__(self, name=None, num=None):
super(TSNResNet, self).__init__()
self.convbn = convbn(3, 16)
self.convpools = dygraph.Sequential(convpool(16, 32, pooling=4),
convpool(32, 64, pooling=4),
convpool(64, 128))
self.fcs = dygraph.Sequential(
dygraph.Linear(7 * 7 * 128, 1024, act='relu'),
dygraph.BatchNorm(1024), dygraph.Dropout(0.5),
dygraph.Linear(1024, 101, act='softmax'))
self.seg_num = 32
def get_activation_norm_layer(num_features, norm_type, input_dim, **norm_params):
"""
Return an activation normalization layer.
"""
input_dim = max(input_dim, 1)
assert input_dim == 2 or input_dim == 1, 'Only support for 2D currently'
if norm_type == 'none' or norm_type == '':
norm_layer = None
elif norm_type == 'batch':
norm_layer = dg.BatchNorm(num_features, **norm_params)
elif norm_type == 'instance':
affine = norm_params.pop('affine', True)
if not affine:
norm_params['param_attr'] = False
norm_params['bias_attr'] = False
norm_layer = dg.InstanceNorm(num_features, **norm_params) #affine=affine, **norm_params)
elif norm_type == 'sync_batch':
affine = norm_params.pop('affine', True)
norm_layer = dg.BatchNorm(num_features, **norm_params)
F.BuildStrategy().sync_batch_norm = True
elif norm_type == 'layer':
norm_layer = dg.LayerNorm(num_features, **norm_params)
elif norm_type == 'layer_2d':
raise NotImplementedError()
elif norm_type == 'adaptive':
norm_layer = AdaptiveNorm(num_features, **norm_params)
elif norm_type == 'spatially_adaptive':
if input_dim != 2:
raise ValueError("Spatially adaptive normalization layers only supports 2D input")
norm_layer = SpatiallyAdaptiveNorm(num_features, **norm_params)
elif norm_type == 'hyper_spatially_adaptive':
if input_dim != 2:
raise ValueError("Spatially adaptive normalization layers only supports 2D input")
norm_layer = HyperSpatiallyAdaptiveNorm(num_features, **norm_params)
else:
raise ValueError("Activation norm layer %s is not recognized" % norm_type)
return norm_layer
def __init__(self,
in_features,
out_features,
groups=1,
kernel_size=3,
padding=1):
super(SameBlock2d, self).__init__()
self.conv = dygraph.Conv2D(num_channels=in_features,
num_filters=out_features,
filter_size=kernel_size,
padding=padding,
groups=groups)
self.norm = dygraph.BatchNorm(out_features)
def convbn(in_channel,
out_channel,
padding=None,
stride=1,
kernel=3,
act=None):
if padding == None:
padding = int((kernel - 1) / 2)
return fluid.dygraph.Sequential(
dygraph.Conv2D(in_channel,
out_channel,
kernel,
stride=stride,
padding=padding,
act=act), dygraph.BatchNorm(out_channel))
def __init__(self):
super(HarFcn, self).__init__()
self.cnn1 = dy.Sequential(
dy.Conv2D(num_channels=1,
num_filters=128,
filter_size=3,
stride=1,
padding=1),
dy.BatchNorm(num_channels=128),
dy.Dropout(p=.2),
)
self.cnn2 = dy.Sequential(
dy.Conv2D(num_channels=128,
num_filters=128,
filter_size=3,
stride=1,
padding=1),
dy.BatchNorm(num_channels=128),
dy.Dropout(p=.2),
)
self.cnn3 = dy.Sequential(
dy.Conv2D(num_channels=128,
num_filters=128,
filter_size=3,
stride=1,
padding=1),
dy.BatchNorm(num_channels=128),
dy.Dropout(p=.2),
)
self.cls = dy.Sequential(
dy.Linear(input_dim=384, output_dim=128),
dy.Dropout(p=.2),
dy.Linear(input_dim=128, output_dim=5),
)
def __init__(self,
in_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1,
act='relu'):
super().__init__()
self.conv = dg.Conv2D(num_channels=in_channels,
num_filters=out_channels,
filter_size=kernel_size,
stride=stride,
padding=padding,
bias_attr=False)
self.bn = dg.BatchNorm(num_channels=out_channels, act=act)
def __init__(self,
in_features,
out_features,
kernel_size=3,
padding=1,
groups=1):
super(DownBlock2d, self).__init__()
self.conv = dygraph.Conv2D(num_channels=in_features,
num_filters=out_features,
filter_size=kernel_size,
padding=padding,
groups=groups)
self.norm = dygraph.BatchNorm(num_channels=out_features, momentum=0.1)
self.pool = dygraph.Pool2D(pool_size=(2, 2),
pool_type='avg',
pool_stride=2)
def convpool(in_channel,
out_channel,
padding=None,
pooling=2,
kernel=3,
act='relu'):
if padding == None:
padding = int((kernel - 1) / 2)
layers = [
dygraph.Conv2D(in_channel,
out_channel,
kernel,
padding=padding,
act=act),
dygraph.BatchNorm(out_channel)
]
if pooling > 1:
layers.append(dygraph.Pool2D(pooling, pool_stride=pooling))
return fluid.dygraph.Sequential(*layers)
def __init__(self, n):
super(FrozenBatchNorm2d, self).__init__()
self.module = dg.BatchNorm(n,
use_global_stats=True,
param_attr=F.ParamAttr(learning_rate=0))
def __init__(self, embedding_size, num_hidden, use_cudnn=True):
""" Encoder prenet layer of TransformerTTS.
Args:
embedding_size (int): the size of embedding.
num_hidden (int): the size of hidden layer in network.
use_cudnn (bool, optional): use cudnn or not. Defaults to True.
"""
super(EncoderPrenet, self).__init__()
self.embedding_size = embedding_size
self.num_hidden = num_hidden
self.use_cudnn = use_cudnn
self.embedding = dg.Embedding(
size=[len(symbols), embedding_size],
padding_idx=0,
param_attr=fluid.initializer.Normal(
loc=0.0, scale=1.0))
self.conv_list = []
k = math.sqrt(1.0 / embedding_size)
self.conv_list.append(
Conv1D(
num_channels=embedding_size,
num_filters=num_hidden,
filter_size=5,
padding=int(np.floor(5 / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(
low=-k, high=k)),
use_cudnn=use_cudnn))
k = math.sqrt(1.0 / num_hidden)
for _ in range(2):
self.conv_list.append(
Conv1D(
num_channels=num_hidden,
num_filters=num_hidden,
filter_size=5,
padding=int(np.floor(5 / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(
low=-k, high=k)),
use_cudnn=use_cudnn))
for i, layer in enumerate(self.conv_list):
self.add_sublayer("conv_list_{}".format(i), layer)
self.batch_norm_list = [
dg.BatchNorm(
num_hidden, data_layout='NCHW') for _ in range(3)
]
for i, layer in enumerate(self.batch_norm_list):
self.add_sublayer("batch_norm_list_{}".format(i), layer)
k = math.sqrt(1.0 / num_hidden)
self.projection = dg.Linear(
num_hidden,
num_hidden,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
def __init__(self,
n_mels=80,
num_hidden=512,
filter_size=5,
padding=0,
num_conv=5,
outputs_per_step=1,
use_cudnn=True,
dropout=0.1,
batchnorm_last=False):
"""Decocder post conv net of TransformerTTS.
Args:
n_mels (int, optional): the number of mel bands when calculating mel spectrograms. Defaults to 80.
num_hidden (int, optional): the size of hidden layer in network. Defaults to 512.
filter_size (int, optional): the filter size of Conv. Defaults to 5.
padding (int, optional): the padding size of Conv. Defaults to 0.
num_conv (int, optional): the num of Conv layers in network. Defaults to 5.
outputs_per_step (int, optional): the num of output frames per step . Defaults to 1.
use_cudnn (bool, optional): use cudnn in Conv or not. Defaults to True.
dropout (float, optional): dropout probability. Defaults to 0.1.
batchnorm_last (bool, optional): if batchnorm at last layer or not. Defaults to False.
"""
super(PostConvNet, self).__init__()
self.dropout = dropout
self.num_conv = num_conv
self.batchnorm_last = batchnorm_last
self.conv_list = []
k = math.sqrt(1.0 / (n_mels * outputs_per_step))
self.conv_list.append(
Conv1D(num_channels=n_mels * outputs_per_step,
num_filters=num_hidden,
filter_size=filter_size,
padding=padding,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(low=-k, high=k)),
use_cudnn=use_cudnn))
k = math.sqrt(1.0 / num_hidden)
for _ in range(1, num_conv - 1):
self.conv_list.append(
Conv1D(
num_channels=num_hidden,
num_filters=num_hidden,
filter_size=filter_size,
padding=padding,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(low=-k, high=k)),
use_cudnn=use_cudnn))
self.conv_list.append(
Conv1D(num_channels=num_hidden,
num_filters=n_mels * outputs_per_step,
filter_size=filter_size,
padding=padding,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(low=-k, high=k)),
use_cudnn=use_cudnn))
for i, layer in enumerate(self.conv_list):
self.add_sublayer("conv_list_{}".format(i), layer)
self.batch_norm_list = [
dg.BatchNorm(num_hidden, data_layout='NCHW')
for _ in range(num_conv - 1)
]
if self.batchnorm_last:
self.batch_norm_list.append(
dg.BatchNorm(n_mels * outputs_per_step, data_layout='NCHW'))
for i, layer in enumerate(self.batch_norm_list):
self.add_sublayer("batch_norm_list_{}".format(i), layer)
def __init__(self,
hidden_size,
batch_size,
K=16,
projection_size=256,
num_gru_layers=2,
max_pool_kernel_size=2,
is_post=False):
"""CBHG Module
Args:
hidden_size (int): dimension of hidden unit.
batch_size (int): batch size of input.
K (int, optional): number of convolution banks. Defaults to 16.
projection_size (int, optional): dimension of projection unit. Defaults to 256.
num_gru_layers (int, optional): number of layers of GRUcell. Defaults to 2.
max_pool_kernel_size (int, optional): max pooling kernel size. Defaults to 2
is_post (bool, optional): whether post processing or not. Defaults to False.
"""
super(CBHG, self).__init__()
self.hidden_size = hidden_size
self.projection_size = projection_size
self.conv_list = []
k = math.sqrt(1.0 / projection_size)
self.conv_list.append(
Conv1D(
num_channels=projection_size,
num_filters=hidden_size,
filter_size=1,
padding=int(np.floor(1 / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(
low=-k, high=k))))
k = math.sqrt(1.0 / hidden_size)
for i in range(2, K + 1):
self.conv_list.append(
Conv1D(
num_channels=hidden_size,
num_filters=hidden_size,
filter_size=i,
padding=int(np.floor(i / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(
initializer=fluid.initializer.Uniform(
low=-k, high=k))))
for i, layer in enumerate(self.conv_list):
self.add_sublayer("conv_list_{}".format(i), layer)
self.batchnorm_list = []
for i in range(K):
self.batchnorm_list.append(
dg.BatchNorm(
hidden_size, data_layout='NCHW'))
for i, layer in enumerate(self.batchnorm_list):
self.add_sublayer("batchnorm_list_{}".format(i), layer)
conv_outdim = hidden_size * K
k = math.sqrt(1.0 / conv_outdim)
self.conv_projection_1 = Conv1D(
num_channels=conv_outdim,
num_filters=hidden_size,
filter_size=3,
padding=int(np.floor(3 / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
k = math.sqrt(1.0 / hidden_size)
self.conv_projection_2 = Conv1D(
num_channels=hidden_size,
num_filters=projection_size,
filter_size=3,
padding=int(np.floor(3 / 2)),
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
self.batchnorm_proj_1 = dg.BatchNorm(hidden_size, data_layout='NCHW')
self.batchnorm_proj_2 = dg.BatchNorm(
projection_size, data_layout='NCHW')
self.max_pool = Pool1D(
pool_size=max_pool_kernel_size,
pool_type='max',
pool_stride=1,
pool_padding=1,
data_format="NCT")
self.highway = Highwaynet(self.projection_size)
h_0 = np.zeros((batch_size, hidden_size // 2), dtype="float32")
h_0 = dg.to_variable(h_0)
k = math.sqrt(1.0 / hidden_size)
self.fc_forward1 = dg.Linear(
hidden_size,
hidden_size // 2 * 3,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
self.fc_reverse1 = dg.Linear(
hidden_size,
hidden_size // 2 * 3,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
self.gru_forward1 = DynamicGRU(
size=self.hidden_size // 2,
is_reverse=False,
origin_mode=True,
h_0=h_0)
self.gru_reverse1 = DynamicGRU(
size=self.hidden_size // 2,
is_reverse=True,
origin_mode=True,
h_0=h_0)
self.fc_forward2 = dg.Linear(
hidden_size,
hidden_size // 2 * 3,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
self.fc_reverse2 = dg.Linear(
hidden_size,
hidden_size // 2 * 3,
param_attr=fluid.ParamAttr(
initializer=fluid.initializer.XavierInitializer()),
bias_attr=fluid.ParamAttr(initializer=fluid.initializer.Uniform(
low=-k, high=k)))
self.gru_forward2 = DynamicGRU(
size=self.hidden_size // 2,
is_reverse=False,
origin_mode=True,
h_0=h_0)
self.gru_reverse2 = DynamicGRU(
size=self.hidden_size // 2,
is_reverse=True,
origin_mode=True,
h_0=h_0)
