espnet.nets.pytorch_backend.transducer.blocks.build_input_layer

source

espnet.nets.pytorch_backend.transducer.blocks.build_input_layer(block: Dict[str, Any], pos_enc_class: Module, padding_idx: int) → Tuple[Conv2dSubsampling | VGG2L | Sequential, int]

Build input layer.

• Parameters:
  • block – Architecture definition of input layer.
  • pos_enc_class – Positional encoding class.
  • padding_idx – Padding symbol ID for embedding layer (if provided).
• Returns: Input layer module. subsampling_factor: Subsampling factor.