espnet2.gan_tts.wavenet.residual_block.Conv1d

source

class espnet2.gan_tts.wavenet.residual_block.Conv1d(*args, **kwargs)

Bases: Conv1d

Conv1d module with customized initialization.

This class is a subclass of torch.nn.Conv1d and provides a customized initialization for the convolutional layer weights using Kaiming normal initialization and sets the bias to zero if it exists.

weight

The learnable weights of the module of shape (out_channels, in_channels, kernel_size).

• Type: Tensor

bias

The learnable bias of the module of shape (out_channels).

• Type: Tensor, optional

• Parameters:

  • *args – Variable length argument list for the parent torch.nn.Conv1d.
  • **kwargs – Arbitrary keyword arguments for the parent torch.nn.Conv1d.

####### Examples

>>> conv = Conv1d(in_channels=3, out_channels=2, kernel_size=5)
>>> print(conv)

Initialize Conv1d module.

reset_parameters()

Reset parameters of the Conv1d layer.

This method initializes the weight of the convolutional layer using the Kaiming normal initialization, which is suitable for layers followed by a ReLU activation function. If the layer has a bias term, it is set to zero.

weight

The weight of the convolutional layer.

• Type: torch.Tensor

bias

The bias of the convolutional layer, if applicable.

• Type: Optional[torch.Tensor]

• Raises:ValueError – If the weight tensor is not initialized properly.

####### Examples

>>> conv_layer = Conv1d(in_channels=1, out_channels=2, kernel_size=3)
>>> conv_layer.reset_parameters()
>>> print(conv_layer.weight)
tensor([[...], ...])  # Initialized weights
>>> print(conv_layer.bias)
tensor([0.])  # Bias initialized to zero