braindecode.modules.SEANetEncoder

class braindecode.modules.SEANetEncoder(dim, n_filters, ratios, kernel_size, last_kernel_size, in_channels=1, n_residual_layers=1, dilation_base=2, compress=2, true_skip=False, causal=False)

EnCodec SEANet encoder: (N, in_channels, L) -> (N, dim, L / prod(ratios)).

Parameters:

• dim (int) – Output channels.

• n_filters (int) – Channels after the first convolution; each downsampling doubles them.

• ratios (sequence of int) – Downsampling ratios, applied in reverse order (EnCodec’s convention).

• kernel_size (int) – Kernel size of the first convolution.

• last_kernel_size (int) – Kernel size of the last convolution.

• in_channels (int) – Input channels.

• n_residual_layers (int) – Residual blocks before each downsampling; block j has dilation dilation_base ** j.

• dilation_base (int) – See n_residual_layers.

• compress (int) – See SEANetResBlock.

• true_skip (bool) – See SEANetResBlock.

• causal (bool) – Left padding only, in every convolution.