espnet2.asr_transducer.encoder.building.build_input_block

source

espnet2.asr_transducer.encoder.building.build_input_block(input_size: int, configuration: Dict[str, int | str]) → ConvInput

Build encoder input block.

This function constructs the input block for the encoder, which typically includes a convolutional layer for processing input features. The configuration dictates the specifics of the convolutional layer, such as size and subsampling factors.

• Parameters:
  • input_size – The size of the input features.
  • configuration – A dictionary containing the input block configuration, which must include the following keys:
    • ‘conv_size’: Size of the convolutional layer.
    • ‘subsampling_factor’: Factor by which to subsample the input.
    • ‘vgg_like’: Boolean indicating whether to use VGG-like architecture.
    • ‘output_size’: Size of the output features after the input block.
• Returns: An instance of the ConvInput block configured as specified.
• Return type:ConvInput

Examples

>>> config = {
...     'conv_size': 3,
...     'subsampling_factor': 2,
...     'vgg_like': True,
...     'output_size': 128
... }
>>> input_block = build_input_block(256, config)
>>> print(input_block)