espnet.nets.pytorch_backend.transducer.blocks.get_pos_enc_and_att_class

source

espnet.nets.pytorch_backend.transducer.blocks.get_pos_enc_and_att_class(net_part: str, pos_enc_type: str, self_attn_type: str) → Tuple[PositionalEncoding | ScaledPositionalEncoding | RelPositionalEncoding, MultiHeadedAttention | RelPositionMultiHeadedAttention]

Get positional encoding and self attention module class.

• Parameters:
  • net_part – Network part, either ‘encoder’ or ‘decoder’.
  • pos_enc_type – Positional encoding type.
  • self_attn_type – Self-attention type.
• Returns: Positional encoding class. self_attn_class: Self-attention class.
• Return type: pos_enc_class