espnet2.gan_codec.shared.discriminator.stft_discriminator.ComplexSTFTResidualUnit

source

espnet2.gan_codec.shared.discriminator.stft_discriminator.ComplexSTFTResidualUnit(in_channel, out_channel, strides)

Complex STFT Residual block for building a discriminator in a GAN framework.

This module serves as a building block for the ComplexSTFTDiscriminator, which processes complex-valued inputs and applies a series of complex convolutions and non-linear activations.

espnet2.gan_codec.shared.discriminator.stft_discriminator.in_channel

Number of input channels for the convolutions.

• Type: int

espnet2.gan_codec.shared.discriminator.stft_discriminator.out_channel

Number of output channels after convolutions.

• Type: int

espnet2.gan_codec.shared.discriminator.stft_discriminator.strides

Stride size for convolutions.

• Type: int

• Parameters:

  • in_channel (int) – Input channel size.
  • out_channel (int) – Output channel size.
  • strides (int) – Stride size of the convolutions.

• Returns: A sequential module containing complex convolutions and ModReLU activation.

• Return type: nn.Module

Examples

>>> complex_stft_unit = ComplexSTFTResidualUnit(1, 2, (1, 2))
>>> output_module = complex_stft_unit(torch.randn(1, 1, 64, 64))
>>> print(output_module.shape)
torch.Size([1, 2, 32, 32])