Method __init__

wenet/efficient_conformer/convolution.py:26–91

Construct an ConvolutionModule object. Args: channels (int): The number of channels of conv layers. kernel_size (int): Kernel size of conv layers. causal (int): Whether use causal convolution or not stride (int): Stride Convolution, for efficie

(self,
                 channels: int,
                 kernel_size: int = 15,
                 activation: nn.Module = nn.ReLU(),
                 norm: str = "batch_norm",
                 causal: bool = False,
                 bias: bool = True,
                 stride: int = 1)

    """ConvolutionModule in Conformer model."""

    def __init__(self,
                 channels: int,
                 kernel_size: int = 15,
                 activation: nn.Module = nn.ReLU(),
                 norm: str = "batch_norm",
                 causal: bool = False,
                 bias: bool = True,
                 stride: int = 1):
        """Construct an ConvolutionModule object.
        Args:
            channels (int): The number of channels of conv layers.
            kernel_size (int): Kernel size of conv layers.
            causal (int): Whether use causal convolution or not
            stride (int): Stride Convolution, for efficient Conformer
        """
        super().__init__()

        self.pointwise_conv1 = nn.Conv1d(
            channels,
            2 * channels,
            kernel_size=1,
            stride=1,
            padding=0,
            bias=bias,
        )
        # self.lorder is used to distinguish if it's a causal convolution,
        # if self.lorder > 0: it's a causal convolution, the input will be
        #    padded with self.lorder frames on the left in forward.
        # else: it's a symmetrical convolution
        if causal:
            padding = 0
            self.lorder = kernel_size - 1
        else:
            # kernel_size should be an odd number for none causal convolution
            assert (kernel_size - 1) % 2 == 0
            padding = (kernel_size - 1) // 2
            self.lorder = 0

        self.depthwise_conv = nn.Conv1d(
            channels,
            channels,
            kernel_size,
            stride=stride,  # for depthwise_conv in StrideConv
            padding=padding,
            groups=channels,
            bias=bias,
        )

        assert norm in ['batch_norm', 'layer_norm']
        if norm == "batch_norm":
            self.use_layer_norm = False
            self.norm = nn.BatchNorm1d(channels)
        else:
            self.use_layer_norm = True
            self.norm = nn.LayerNorm(channels)

        self.pointwise_conv2 = nn.Conv1d(
            channels,