Class Conv1dSubsampling

funasr/models/transformer/utils/subsampling.py:332–388

Convolutional 1D subsampling (to 1/2 length). Args: idim (int): Input dimension. odim (int): Output dimension. dropout_rate (float): Dropout rate. pos_enc (torch.nn.Module): Custom position encoding layer.

class Conv1dSubsampling(torch.nn.Module):
    """Convolutional 1D subsampling (to 1/2 length).

    Args:
        idim (int): Input dimension.
        odim (int): Output dimension.
        dropout_rate (float): Dropout rate.
        pos_enc (torch.nn.Module): Custom position encoding layer.

    """

    def __init__(
        self,
        idim,
        odim,
        kernel_size,
        stride,
        pad,
        tf2torch_tensor_name_prefix_torch: str = "stride_conv",
        tf2torch_tensor_name_prefix_tf: str = "seq2seq/proj_encoder/downsampling",
    ):
        """Initialize Conv1dSubsampling.
        
            Args:
                idim: TODO.
                odim: TODO.
                kernel_size: Size/dimension parameter.
                stride: TODO.
                pad: TODO.
                tf2torch_tensor_name_prefix_torch: TODO.
                tf2torch_tensor_name_prefix_tf: TODO.
            """
        super(Conv1dSubsampling, self).__init__()
        self.conv = torch.nn.Conv1d(idim, odim, kernel_size, stride)
        self.pad_fn = torch.nn.ConstantPad1d(pad, 0.0)
        self.stride = stride
        self.odim = odim
        self.tf2torch_tensor_name_prefix_torch = tf2torch_tensor_name_prefix_torch
        self.tf2torch_tensor_name_prefix_tf = tf2torch_tensor_name_prefix_tf

    def output_size(self) -> int:
        """Output size."""
        return self.odim

    def forward(self, x, x_len):
        """Subsample x."""
        x = x.transpose(1, 2)  # (b, d ,t)
        x = self.pad_fn(x)
        # x = F.relu(self.conv(x))
        x = F.leaky_relu(self.conv(x), negative_slope=0.0)
        x = x.transpose(1, 2)  # (b, t ,d)

        if x_len is None:

            return x, None
        x_len = (x_len - 1) // self.stride + 1
        return x, x_len