espnet2.legacy.nets.pytorch_backend.conformer.convolution.mask_padded_frames

source

espnet2.legacy.nets.pytorch_backend.conformer.convolution.mask_padded_frames(x, mask_pad)

Zero the padded time steps of x (#batch, channels, time).

mask_pad is (#batch, 1, time) with True/1 on real frames, or None. A mask whose time axis does not match x (the single-frame query of a cached step, for instance) is ignored, since it cannot be applied.