espnet2.asr_transducer.encoder.building.build_conv1d_block

source

espnet2.asr_transducer.encoder.building.build_conv1d_block(configuration: List[Dict[str, Any]], causal: bool) → Conv1d

Build Conv1d block.

This function constructs a Conv1d block based on the provided configuration and causal flag. The Conv1d block is used in various neural network architectures for processing sequential data.

• Parameters:

  • configuration –

    A list of dictionaries where each dictionary contains the configuration parameters for the Conv1d block. Expected keys include:

    • input_size: Size of the input features.
    • output_size: Size of the output features.
    • kernel_size: Size of the convolutional kernel.
    • stride: Stride of the convolution (default is 1).
    • dilation: Dilation factor (default is 1).
    • groups: Number of groups for grouped convolution

    (default is 1).

    • bias: Whether to include a bias term (default is True).
    • relu: Whether to apply ReLU activation (default is True).
    • batch_norm: Whether to include batch normalization : (default is False).

  • causal – A boolean indicating whether the convolution should be causal. If True, the convolution will not include future time steps.

• Returns: A Conv1d block function that can be called to create the Conv1d layer.

Examples

>>> conv1d_block = build_conv1d_block(
...     configuration=[
...         {
...             "input_size": 128,
...             "output_size": 256,
...             "kernel_size": 3,
...             "stride": 1,
...             "dilation": 1,
...             "groups": 1,
...             "bias": True,
...             "relu": True,
...             "batch_norm": False,
...         }
...     ],
...     causal=True
... )
>>> conv_layer = conv1d_block()