espnet2.enh.layers.adapt_layers.make_adapt_layer

source

espnet2.enh.layers.adapt_layers.make_adapt_layer(type, indim, enrolldim, ninputs=1)

Creates an adaptation layer based on the specified type.

This function returns an instance of an adaptation layer class defined in the adaptation_layer_types dictionary. The type of layer to create is specified by the type argument. The input dimensions and enrollment dimensions must also be provided.

espnet2.enh.layers.adapt_layers.type

The type of adaptation layer to create. Possible values are “concat”, “muladd”, and “mul”.

• Type: str

espnet2.enh.layers.adapt_layers.indim

The input dimension for the adaptation layer.

• Type: int

espnet2.enh.layers.adapt_layers.enrolldim

The enrollment dimension for the adaptation layer.

• Type: int

espnet2.enh.layers.adapt_layers.ninputs

The number of input tensors to the layer. Defaults to 1.

• Type: int, optional

• Parameters:

  • type (str) – The type of adaptation layer to create.
  • indim (int) – The input dimension for the adaptation layer.
  • enrolldim (int) – The enrollment dimension for the adaptation layer.
  • ninputs (int , optional) – The number of input tensors to the layer. Defaults to 1.

• Returns: An instance of the specified adaptation layer class.

• Return type: nn.Module

• Raises:KeyError – If the specified type does not exist in adaptation_layer_types.

Examples

>>> layer = make_adapt_layer("concat", 128, 256)
>>> print(layer)
ConcatAdaptLayer(...)

>>> layer = make_adapt_layer("muladd", 128, 256, ninputs=2)
>>> print(layer)
MulAddAdaptLayer(...)