espnet.nets.pytorch_backend.transformer.embedding.StreamPositionalEncoding

source

class espnet.nets.pytorch_backend.transformer.embedding.StreamPositionalEncoding(d_model, dropout_rate, max_len=5000)

Bases: Module

Streaming Positional encoding.

• Parameters:
  • d_model (int) – Embedding dimension.
  • dropout_rate (float) – Dropout rate.
  • max_len (int) – Maximum input length.

Construct an PositionalEncoding object.

extend_pe(length, device, dtype)

Reset the positional encodings.

forward(x: Tensor, start_idx: int = 0)

Add positional encoding.

• Parameters:x (torch.Tensor) – Input tensor (batch, time, *).
• Returns: Encoded tensor (batch, time, *).
• Return type: torch.Tensor