def __init__(self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=10,
conv_channels=64,
dilation_factor=1,
nonlinear_activation="LeakyReLU",
nonlinear_activation_params={"negative_slope": 0.2},
bias=True,
use_weight_norm=True,
):
"""Initialize Parallel WaveGAN Discriminator module.
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
kernel_size (int): Number of output channels.
layers (int): Number of conv layers.
conv_channels (int): Number of chnn layers.
dilation_factor (int): Dilation factor. For example, if dilation_factor = 2,
the dilation will be 2, 4, 8, ..., and so on.
nonlinear_activation (str): Nonlinear function after each conv.
nonlinear_activation_params (dict): Nonlinear function parameters
bias (bool): Whether to use bias parameter in conv.
use_weight_norm (bool) Whether to use weight norm.
If set to true, it will be applied to all of the conv layers.
"""
super(ParallelWaveGANDiscriminator, self).__init__()
assert (kernel_size - 1) % 2 == 0, "Not support even number kernel size."
assert dilation_factor > 0, "Dilation factor must be > 0."
self.conv_layers = torch.nn.ModuleList()
conv_in_channels = in_channels
for i in range(layers - 1):
if i == 0:
dilation = 1
else:
dilation = i if dilation_factor == 1 else dilation_factor ** i
conv_in_channels = conv_channels
padding = (kernel_size - 1) // 2 * dilation
conv_layer = [
Conv1d(conv_in_channels, conv_channels,
kernel_size=kernel_size, padding=padding,
dilation=dilation, bias=bias),
getattr(torch.nn, nonlinear_activation)(inplace=True, **nonlinear_activation_params)
]
self.conv_layers += conv_layer
padding = (kernel_size - 1) // 2
last_conv_layer = Conv1d(
conv_in_channels, out_channels,
kernel_size=kernel_size, padding=padding, bias=bias)
self.conv_layers += [last_conv_layer]
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
def __init__(self,
upsample_scales,
upsample_activation="none",
upsample_activation_params={},
mode="nearest",
freq_axis_kernel_size=1,
aux_channels=80,
aux_context_window=0):
"""Initialize convolution + upsampling network module.
Args:
upsample_scales (list): List of upsampling scales.
upsample_activation (str): Activation function name.
upsample_activation_params (dict): Arguments for specified activation function.
mode (str): Interpolation mode.
freq_axis_kernel_size (int): Kernel size in the direction of frequency axis.
aux_channels (int): Number of channels of pre-convolutional layer.
aux_context_window (int): Context window size of the pre-convolutional layer.
"""
super(ConvInUpsampleNetwork, self).__init__()
# To capture wide-context information in conditional features
kernel_size = 2 * aux_context_window + 1
# NOTE(kan-bayashi): Here do not use padding because the input is already padded
self.conv_in = Conv1d(aux_channels,
aux_channels,
kernel_size=kernel_size,
bias=False)
self.upsample = UpsampleNetwork(
upsample_scales=upsample_scales,
upsample_activation=upsample_activation,
upsample_activation_params=upsample_activation_params,
mode=mode,
freq_axis_kernel_size=freq_axis_kernel_size,
)
def __init__(self,
in_channels=1,
out_channels=1,
kernel_size=3,
layers=10,
conv_channels=64,
dilation_factor=1,
nonlinear_activation="LeakyReLU",
nonlinear_activation_params={"negative_slope": 0.2},
bias=True,
use_weight_norm=True,
):
super(ParallelWaveGANDiscriminator, self).__init__()
assert (kernel_size - 1) % 2 == 0, "Not support even number kernel size."
assert dilation_factor > 0, "Dilation factor must be > 0."
self.conv_layers = torch.nn.ModuleList()
conv_in_channels = in_channels
for i in range(layers - 1):
if i == 0:
dilation = 1
else:
dilation = i if dilation_factor == 1 else dilation_factor ** i
conv_in_channels = conv_channels
padding = (kernel_size - 1) // 2 * dilation
conv_layer = [
Conv1d(conv_in_channels, conv_channels,
kernel_size=kernel_size, padding=padding,
dilation=dilation, bias=bias),
getattr(torch.nn, nonlinear_activation)(inplace=True, **nonlinear_activation_params)
]
self.conv_layers += conv_layer
padding = (kernel_size - 1) // 2
last_conv_layer = Conv1d(
conv_in_channels, out_channels,
kernel_size=kernel_size, padding=padding, bias=bias)
self.conv_layers += [last_conv_layer]
# apply weight norm
if use_weight_norm:
self.apply_weight_norm()
def __init__(self, upsample_scales, nonlinear_activation=None, nonlinear_activation_params={}, interpolate_mode="nearest", freq_axis_kernel_size=1,
aux_channels=80, aux_context_window=0, use_causal_conv=False ):
super(ConvInUpsampleNetwork, self).__init__()
self.aux_context_window = aux_context_window
self.use_causal_conv = use_causal_conv and aux_context_window > 0
# To capture wide-context information in conditional features
kernel_size = aux_context_window + 1 if use_causal_conv else 2 * aux_context_window + 1
self.conv_in = Conv1d(aux_channels, aux_channels, kernel_size=kernel_size, bias=False)
self.upsample = UpsampleNetwork( upsample_scales=upsample_scales, nonlinear_activation=nonlinear_activation, nonlinear_activation_params=nonlinear_activation_params,
interpolate_mode=interpolate_mode, freq_axis_kernel_size=freq_axis_kernel_size, use_causal_conv=use_causal_conv,)
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,
upsample_scales,
nonlinear_activation=None,
nonlinear_activation_params={},
interpolate_mode="nearest",
freq_axis_kernel_size=1,
aux_channels=80,
aux_context_window=2,
aux_context_pad=True,
use_causal_conv=False
):
"""Initialize convolution + upsampling network module.
Args:
upsample_scales (list): List of upsampling scales.
nonlinear_activation (str): Activation function name.
nonlinear_activation_params (dict): Arguments for specified activation function.
mode (str): Interpolation mode.
freq_axis_kernel_size (int): Kernel size in the direction of frequency axis.
aux_channels (int): Number of channels of pre-convolutional layer.
aux_context_window (int): Context window size of the pre-convolutional layer.
use_causal_conv (bool): Whether to use causal structure.
"""
super(ConvInUpsampleNetwork, self).__init__()
self.aux_context_window = aux_context_window
self.use_causal_conv = use_causal_conv and aux_context_window > 0
# To capture wide-context information in conditional features
kernel_size = aux_context_window + 1 if use_causal_conv else 2 * aux_context_window + 1
# NOTE(kan-bayashi): Here do not use padding because the input is already padded
padd = aux_context_window if aux_context_pad else 0
self.conv_in = Conv1d(aux_channels, aux_channels, kernel_size=kernel_size, padding=padd, bias=False)
self.upsample = UpsampleNetwork(
upsample_scales=upsample_scales,
nonlinear_activation=nonlinear_activation,
nonlinear_activation_params=nonlinear_activation_params,
interpolate_mode=interpolate_mode,
freq_axis_kernel_size=freq_axis_kernel_size,
use_causal_conv=use_causal_conv,
)
