espnet.nets.pytorch_backend.transducer.rnn_encoder.VGG2L

source

class espnet.nets.pytorch_backend.transducer.rnn_encoder.VGG2L(in_channel: int = 1)

Bases: Module

VGG-like module.

• Parameters:in_channel – number of input channels

Initialize VGG-like module.

forward(feats: Tensor, feats_len: Tensor, **kwargs) → Tuple[Tensor, Tensor, Tensor | None]

VGG2L forward.

• Parameters:
  • feats – Feature sequences. (B, F, D_feats)
  • feats_len – Feature sequences lengths. (B, )
• Returns: VGG2L output sequences. (B, F // 4, 128 * D_feats // 4) vgg_out_len: VGG2L output sequences lengths. (B,)
• Return type: vgg_out