def __init__(self,in_channels=1, out_channels=1,kernel_size=3,layers=30,stacks=3,residual_channels=64,gate_channels=128,skip_channels=64,aux_channels=80,aux_context_window=2,
dropout=0.0,bias=True,use_weight_norm=True,use_causal_conv=False,upsample_conditional_features=True,upsample_net="ConvInUpsampleNetwork", upsample_params={"upsample_scales": [4, 4, 4, 4]},
):
super(ParallelWaveGANGenerator, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.aux_channels = aux_channels
self.layers = layers
self.stacks = stacks
self.kernel_size = kernel_size
# check the number of layers and stacks
assert layers % stacks == 0
layers_per_stack = layers // stacks
# define first convolution
self.first_conv = Conv1d1x1(in_channels, residual_channels, bias=True)
# define conv + upsampling network
if upsample_conditional_features:
upsample_params.update({
"use_causal_conv": use_causal_conv,
})
if upsample_net == "ConvInUpsampleNetwork":
upsample_params.update({
"aux_channels": aux_channels,
"aux_context_window": aux_context_window,
})
self.upsample_net = getattr(upsample, upsample_net)(**upsample_params)
else:
self.upsample_net = None
# define residual blocks
self.conv_layers = torch.nn.ModuleList()
for layer in range(layers):
dilation = 2 ** (layer % layers_per_stack)
conv = ResidualBlock( kernel_size=kernel_size, residual_channels=residual_channels, gate_channels=gate_channels, skip_channels=skip_channels, aux_channels=aux_channels, dilation=dilation,
dropout=dropout,bias=bias, use_causal_conv=use_causal_conv,)
self.conv_layers += [conv]
# define output layers
self.last_conv_layers = torch.nn.ModuleList([
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, skip_channels, bias=True),
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, out_channels, bias=True),
])
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
def test_conv_initialization():
conv = Conv1d(10, 10, 3, bias=True)
np.testing.assert_array_equal(conv.bias.data.numpy(),
np.zeros_like(conv.bias.data.numpy()))
conv1x1 = Conv1d1x1(10, 10, bias=True)
np.testing.assert_array_equal(conv1x1.bias.data.numpy(),
np.zeros_like(conv1x1.bias.data.numpy()))
kernel_size = (10, 10)
conv2d = Conv2d(10, 10, kernel_size, bias=True)
np.testing.assert_array_equal(
conv2d.weight.data.numpy(),
np.ones_like(conv2d.weight.data.numpy()) / np.prod(kernel_size))
np.testing.assert_array_equal(conv2d.bias.data.numpy(),
np.zeros_like(conv2d.bias.data.numpy()))
kernel_size = (1, 10)
conv2d = Conv2d(10, 10, kernel_size, bias=True)
np.testing.assert_array_equal(
conv2d.weight.data.numpy(),
np.ones_like(conv2d.weight.data.numpy()) / np.prod(kernel_size))
np.testing.assert_array_equal(conv2d.bias.data.numpy(),
np.zeros_like(conv2d.bias.data.numpy()))
def __init__(
self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=30,
stacks=3,
residual_channels=64,
gate_channels=128,
skip_channels=64,
dropout=0.0,
use_weight_norm=True,
use_causal_conv=False,
nonlinear_activation_params={"negative_slope": 0.2},
):
"""Initialize Parallel WaveGAN Discriminator module.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_size (int): Kernel size of dilated convolution.
layers (int): Number of residual block layers.
stacks (int): Number of stacks i.e., dilation cycles.
residual_channels (int): Number of channels in residual conv.
gate_channels (int): Number of channels in gated conv.
skip_channels (int): Number of channels in skip conv.
dropout (float): Dropout rate. 0.0 means no dropout applied.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv (bool): Whether to use causal structure.
nonlinear_activation_params (dict): Nonlinear function parameters
"""
super(ResidualParallelWaveGANDiscriminator, self).__init__()
assert (kernel_size -
1) % 2 == 0, "Not support even number kernel size."
self.in_channels = in_channels
self.out_channels = out_channels
self.layers = layers
self.stacks = stacks
self.kernel_size = kernel_size
# check the number of layers and stacks
assert layers % stacks == 0
layers_per_stack = layers // stacks
# define first convolution
self.first_conv = Conv1d1x1(in_channels, residual_channels, bias=True)
self.first_conv_activation = torch.nn.LeakyReLU(
negative_slope=nonlinear_activation_params['negative_slope'],
inplace=True)
# define residual blocks
self.conv_layers = torch.nn.ModuleList()
for layer in range(layers):
dilation = 2**(layer % layers_per_stack)
conv = ResidualBlock(
kernel_size=kernel_size,
residual_channels=residual_channels,
gate_channels=gate_channels,
skip_channels=skip_channels,
aux_channels=-1,
dilation=dilation,
dropout=dropout,
bias=True, # NOTE: magenda uses bias, but musyoku doesn't
use_causal_conv=use_causal_conv,
)
self.conv_layers += [conv]
# define output layers
self.last_conv_layers = torch.nn.ModuleList([
torch.nn.LeakyReLU(
negative_slope=nonlinear_activation_params['negative_slope'],
inplace=True),
Conv1d1x1(skip_channels, skip_channels, bias=True),
torch.nn.LeakyReLU(
negative_slope=nonlinear_activation_params['negative_slope'],
inplace=True),
Conv1d1x1(skip_channels, out_channels, bias=True),
])
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
def __init__(
self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=30,
stacks=3,
residual_channels=64,
gate_channels=128,
skip_channels=64,
aux_channels=80,
aux_context_window=2,
dropout=0.0,
use_weight_norm=True,
use_causal_conv=False,
upsample_conditional_features=True,
upsample_net="ConvInUpsampleNetwork",
upsample_params={"upsample_scales": [4, 4, 4, 4]},
):
"""Initialize Parallel WaveGAN Generator module.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_size (int): Kernel size of dilated convolution.
layers (int): Number of residual block layers.
stacks (int): Number of stacks i.e., dilation cycles.
residual_channels (int): Number of channels in residual conv.
gate_channels (int): Number of channels in gated conv.
skip_channels (int): Number of channels in skip conv.
aux_channels (int): Number of channels for auxiliary feature conv.
aux_context_window (int): Context window size for auxiliary feature.
dropout (float): Dropout rate. 0.0 means no dropout applied.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv (bool): Whether to use causal structure.
upsample_conditional_features (bool): Whether to use upsampling network.
upsample_net (str): Upsampling network architecture.
upsample_params (dict): Upsampling network parameters.
"""
super(ParallelWaveGANGenerator, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.aux_channels = aux_channels
self.layers = layers
self.stacks = stacks
self.kernel_size = kernel_size
# check the number of layers and stacks
assert layers % stacks == 0
layers_per_stack = layers // stacks
# define first convolution
self.first_conv = Conv1d1x1(in_channels, residual_channels, bias=True)
# define conv + upsampling network
if upsample_conditional_features:
upsample_params.update({
"use_causal_conv": use_causal_conv,
})
if upsample_net == "ConvInUpsampleNetwork":
upsample_params.update({
"aux_channels":
aux_channels,
"aux_context_window":
aux_context_window,
})
self.upsample_net = getattr(upsample,
upsample_net)(**upsample_params)
else:
self.upsample_net = None
# define residual blocks
self.conv_layers = torch.nn.ModuleList()
for layer in range(layers):
dilation = 2**(layer % layers_per_stack)
conv = ResidualBlock(
kernel_size=kernel_size,
residual_channels=residual_channels,
gate_channels=gate_channels,
skip_channels=skip_channels,
aux_channels=aux_channels,
dilation=dilation,
dropout=dropout,
bias=True, # NOTE: magenda uses bias, but musyoku doesn't
use_causal_conv=use_causal_conv,
)
self.conv_layers += [conv]
# define output layers
self.last_conv_layers = torch.nn.ModuleList([
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, skip_channels, bias=True),
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, out_channels, bias=True),
])
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
def __init__(self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=30,
stacks=3,
residual_channels=64,
gate_channels=128,
skip_channels=64,
aux_channels=80,
aux_context_window=2,
dropout=0.0,
bias=True,
use_weight_norm=True,
use_causal_conv=False,
upsample_conditional_features=True,
upsample_net="ConvInUpsampleNetwork",
upsample_params={"upsample_scales": [4, 4, 4, 4]}, # 4, 5, 3, 5
):
"""Initialize Parallel WaveGAN Generator module.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_size (int): Kernel size of dilated convolution.
layers (int): Number of residual block layers.
stacks (int): Number of stacks i.e., dilation cycles.
residual_channels (int): Number of channels in residual conv.
gate_channels (int): Number of channels in gated conv.
skip_channels (int): Number of channels in skip conv.
aux_channels (int): Number of channels for auxiliary feature conv.
aux_context_window (int): Context window size for auxiliary feature.
dropout (float): Dropout rate. 0.0 means no dropout applied.
bias (bool): Whether to use bias parameter in conv layer.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv (bool): Whether to use causal structure.
upsample_conditional_features (bool): Whether to use upsampling network.
upsample_net (str): Upsampling network architecture.
upsample_params (dict): Upsampling network parameters.
"""
super(ParallelWaveGANGenerator, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.aux_channels = aux_channels
self.aux_context_window = aux_context_window
self.layers = layers
self.stacks = stacks
self.kernel_size = kernel_size
# check the number of layers and stacks
assert layers % stacks == 0
layers_per_stack = layers // stacks
# define first convolution
# 1 -> 64
# B 64 L
self.first_conv = Conv1d1x1(in_channels, residual_channels, bias=True)
# define conv + upsampling network
if upsample_conditional_features:
upsample_params.update({
"use_causal_conv": use_causal_conv,
})
if upsample_net == "MelGANGenerator":
assert aux_context_window == 0
upsample_params.update({
"use_weight_norm": False, # not to apply twice
"use_final_nonlinear_activation": False,
})
self.upsample_net = getattr(models, upsample_net)(**upsample_params)
else:
if upsample_net == "ConvInUpsampleNetwork":
# 4, 5, 3, 5 upsample scales
# following two params are not used
upsample_params.update({
"aux_channels": aux_channels, # 80
"aux_context_window": aux_context_window, # 2
})
# in conv, 80->80, 2*2win+1 kernel, padded by 2win already: T -> T-2win
# upsample:
# stretch:B 1 80 T -> B 1 80 (T-2win)*scale: 4 5 3 5, nearest neighbor upsampling
# conv2d:1, 1, 9 11 7 11, 4 5 3 5
# no bias
# return B 80 (T-2win)*scale:L
# note: every node will connect to 2 nodes in next layer, and with different distance for adjacent layers, this will not cause the same gradient.
self.upsample_net = getattr(upsample, upsample_net)(**upsample_params)
self.upsample_factor = np.prod(upsample_params["upsample_scales"]) # 300, hop size
else:
self.upsample_net = None
self.upsample_factor = 1
# define residual blocks
self.conv_layers = torch.nn.ModuleList()
for layer in range(layers):
dilation = 2 ** (layer % layers_per_stack)
# 1. conv1d kernel 3, dilated: B 64 L -> B 128 L
# 2. B 128 L -> B 64 L, B 64 L
# 3. B 80 L 1x1 -> B 128 L -> B 64 L, B 64 L
# 4. tanh * sigmoid of the sums
# 5. 1x1 -> skip channel B 64 L
# 6. 1x1 res 1/sqrt(2) -> res channel B 64 L
conv = ResidualBlock(
kernel_size=kernel_size, #1
residual_channels=residual_channels, # 64
gate_channels=gate_channels, # 128
skip_channels=skip_channels, # 64
aux_channels=aux_channels, # 80
dilation=dilation,
dropout=dropout, # 0
bias=bias, # True
use_causal_conv=use_causal_conv, # False
)
self.conv_layers += [conv]
# define output layers
# 64->64->1
self.last_conv_layers = torch.nn.ModuleList([
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, skip_channels, bias=True),
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, out_channels, bias=True),
])
# apply weight norm
if use_weight_norm: # True
self.apply_weight_norm()
def __init__(self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=30,
stacks=3,
residual_channels=64,
gate_channels=128,
skip_channels=64,
aux_channels=80,
aux_context_window=2,
dropout=0.0,
bias=True,
use_weight_norm=True,
use_causal_conv=False,
upsample_conditional_features=True,
upsample_net="ConvInUpsampleNetwork",
upsample_params={"upsample_scales": [4, 4, 4, 4]},
use_asr_layer=False,
asr_config=None,
asr_pretrained_file=None,
asr_feature_layer_nth=-1
):
"""Initialize Parallel WaveGAN Generator module.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_size (int): Kernel size of dilated convolution.
layers (int): Number of residual block layers.
stacks (int): Number of stacks i.e., dilation cycles.
residual_channels (int): Number of channels in residual conv.
gate_channels (int): Number of channels in gated conv.
skip_channels (int): Number of channels in skip conv.
aux_channels (int): Number of channels for auxiliary feature conv.
aux_context_window (int): Context window size for auxiliary feature.
dropout (float): Dropout rate. 0.0 means no dropout applied.
bias (bool): Whether to use bias parameter in conv layer.
use_weight_norm (bool): Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
use_causal_conv (bool): Whether to use causal structure.
upsample_conditional_features (bool): Whether to use upsampling network.
upsample_net (str): Upsampling network architecture.
upsample_params (dict): Upsampling network parameters.
"""
super(ParallelWaveGANGenerator, self).__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.aux_channels = aux_channels
self.aux_context_window = aux_context_window
self.layers = layers
self.stacks = stacks
self.kernel_size = kernel_size
# check the number of layers and stacks
assert layers % stacks == 0
layers_per_stack = layers // stacks
# define first convolution
self.first_conv = Conv1d1x1(in_channels, residual_channels, bias=True)
# define conv + upsampling network
if upsample_conditional_features:
upsample_params.update({
"use_causal_conv": use_causal_conv,
})
if upsample_net == "MelGANGenerator":
assert aux_context_window == 0
upsample_params.update({
"use_weight_norm": False, # not to apply twice
"use_final_nonlinear_activation": False,
})
self.upsample_net = getattr(models, upsample_net)(**upsample_params)
else:
if upsample_net == "ConvInUpsampleNetwork":
upsample_params.update({
"aux_channels": aux_channels,
"aux_context_window": aux_context_window,
})
self.upsample_net = getattr(upsample, upsample_net)(**upsample_params)
self.upsample_factor = np.prod(upsample_params["upsample_scales"])
else:
self.upsample_net = None
self.upsample_factor = 1
# define residual blocks
self.conv_layers = torch.nn.ModuleList()
for layer in range(layers):
dilation = 2 ** (layer % layers_per_stack)
conv = ResidualBlock(
kernel_size=kernel_size,
residual_channels=residual_channels,
gate_channels=gate_channels,
skip_channels=skip_channels,
aux_channels=aux_channels,
dilation=dilation,
dropout=dropout,
bias=bias,
use_causal_conv=use_causal_conv,
)
self.conv_layers += [conv]
# define output layers
self.last_conv_layers = torch.nn.ModuleList([
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, skip_channels, bias=True),
torch.nn.ReLU(inplace=True),
Conv1d1x1(skip_channels, out_channels, bias=True),
])
if use_asr_layer:
conf = OmegaConf.load(asr_config)
encoder_conf = conf.model.encoder
jasper = encoder_conf.jasper
feat_in = encoder_conf.feat_in
activation = encoder_conf.activation
conv_mask = encoder_conf.conv_mask
self.asr_layer = ConvASREncoder(jasper, activation, feat_in, conv_mask=conv_mask)
for m in self.asr_layer.modules():
if isinstance(m, MaskedConv1d):
m.use_mask = False
else:
self.asr_layer = None
self.asr_feature_layer_nth = asr_feature_layer_nth
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
