espnet2.beats.utils.forward_padding_mask_conv

source

espnet2.beats.utils.forward_padding_mask_conv(padding_mask: Tensor, n_dim: int, conv_module: Module)

Forward padding mask.

To be applied after features are passed through conv module or after converting to spectrogram, for consistency. padding_mask: BT n_dim: number of dimensions before the transformation was applied to features.

When applying after fbank computation set this to a non-positive value.

conv_module: conv module applied to features, : the channel dimension must be 1.