espnet2.legacy.nets.pytorch_backend.transducer.joint_network.JointNetwork

source

class espnet2.legacy.nets.pytorch_backend.transducer.joint_network.JointNetwork(joint_output_size: int, encoder_output_size: int, decoder_output_size: int, joint_space_size: int, joint_activation_type: int)

Bases: Module

Transducer joint network module.

• Parameters:
  • joint_output_size – Joint network output dimension
  • encoder_output_size – Encoder output dimension.
  • decoder_output_size – Decoder output dimension.
  • joint_space_size – Dimension of joint space.
  • joint_activation_type – Type of activation for joint network.

Joint network initializer.

forward(enc_out: Tensor, dec_out: Tensor, is_aux: bool = False, quantization: bool = False) → Tensor

Joint computation of encoder and decoder hidden state sequences.

• Parameters:
  • enc_out – Expanded encoder output state sequences (B, T, 1, D_enc)
  • dec_out – Expanded decoder output state sequences (B, 1, U, D_dec)
  • is_aux – Whether auxiliary tasks in used.
  • quantization – Whether dynamic quantization is used.
• Returns: Joint output state sequences. (B, T, U, D_out)
• Return type: joint_out