modules/wenet_extractor/efficient_conformer/encoder_layer.py

## This module is from [WeNet](https://github.com/wenet-e2e/wenet).

# ## Citations

# ```bibtex
# @inproceedings{yao2021wenet,
#   title={WeNet: Production oriented Streaming and Non-streaming End-to-End Speech Recognition Toolkit},
#   author={Yao, Zhuoyuan and Wu, Di and Wang, Xiong and Zhang, Binbin and Yu, Fan and Yang, Chao and Peng, Zhendong and Chen, Xiaoyu and Xie, Lei and Lei, Xin},
#   booktitle={Proc. Interspeech},
#   year={2021},
#   address={Brno, Czech Republic },
#   organization={IEEE}
# }

# @article{zhang2022wenet,
#   title={WeNet 2.0: More Productive End-to-End Speech Recognition Toolkit},
#   author={Zhang, Binbin and Wu, Di and Peng, Zhendong and Song, Xingchen and Yao, Zhuoyuan and Lv, Hang and Xie, Lei and Yang, Chao and Pan, Fuping and Niu, Jianwei},
#   journal={arXiv preprint arXiv:2203.15455},
#   year={2022}
# }
#

"""Encoder self-attention layer definition."""

from typing import Optional, Tuple
import torch
from torch import nn


class StrideConformerEncoderLayer(nn.Module):
    """Encoder layer module.
    Args:
        size (int): Input dimension.
        self_attn (torch.nn.Module): Self-attention module instance.
            `MultiHeadedAttention` or `RelPositionMultiHeadedAttention`
            instance can be used as the argument.
        feed_forward (torch.nn.Module): Feed-forward module instance.
            `PositionwiseFeedForward` instance can be used as the argument.
        feed_forward_macaron (torch.nn.Module): Additional feed-forward module
             instance.
            `PositionwiseFeedForward` instance can be used as the argument.
        conv_module (torch.nn.Module): Convolution module instance.
            `ConvlutionModule` instance can be used as the argument.
        dropout_rate (float): Dropout rate.
        normalize_before (bool):
            True: use layer_norm before each sub-block.
            False: use layer_norm after each sub-block.
    """

    def __init__(
        self,
        size: int,
        self_attn: torch.nn.Module,
        feed_forward: Optional[nn.Module] = None,
        feed_forward_macaron: Optional[nn.Module] = None,
        conv_module: Optional[nn.Module] = None,
        pointwise_conv_layer: Optional[nn.Module] = None,
        dropout_rate: float = 0.1,
        normalize_before: bool = True,
    ):
        """Construct an EncoderLayer object."""
        super().__init__()
        self.self_attn = self_attn
        self.feed_forward = feed_forward
        self.feed_forward_macaron = feed_forward_macaron
        self.conv_module = conv_module
        self.pointwise_conv_layer = pointwise_conv_layer
        self.norm_ff = nn.LayerNorm(size, eps=1e-5)  # for the FNN module
        self.norm_mha = nn.LayerNorm(size, eps=1e-5)  # for the MHA module
        if feed_forward_macaron is not None:
            self.norm_ff_macaron = nn.LayerNorm(size, eps=1e-5)
            self.ff_scale = 0.5
        else:
            self.ff_scale = 1.0
        if self.conv_module is not None:
            self.norm_conv = nn.LayerNorm(size, eps=1e-5)  # for the CNN module
            self.norm_final = nn.LayerNorm(
                size, eps=1e-5
            )  # for the final output of the block
        self.dropout = nn.Dropout(dropout_rate)
        self.size = size
        self.normalize_before = normalize_before
        self.concat_linear = nn.Linear(size + size, size)

    def forward(
        self,
        x: torch.Tensor,
        mask: torch.Tensor,
        pos_emb: torch.Tensor,
        mask_pad: torch.Tensor = torch.ones((0, 0, 0), dtype=torch.bool),
        att_cache: torch.Tensor = torch.zeros((0, 0, 0, 0)),
        cnn_cache: torch.Tensor = torch.zeros((0, 0, 0, 0)),
    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
        """Compute encoded features.

        Args:
            x (torch.Tensor): (#batch, time, size)
            mask (torch.Tensor): Mask tensor for the input (#batch, time，time),
                (0, 0, 0) means fake mask.
            pos_emb (torch.Tensor): positional encoding, must not be None
                for ConformerEncoderLayer.
            mask_pad (torch.Tensor): batch padding mask used for conv module.
                (#batch, 1，time), (0, 0, 0) means fake mask.
            att_cache (torch.Tensor): Cache tensor of the KEY & VALUE
                (#batch=1, head, cache_t1, d_k * 2), head * d_k == size.
            cnn_cache (torch.Tensor): Convolution cache in conformer layer
                (#batch=1, size, cache_t2)
        Returns:
            torch.Tensor: Output tensor (#batch, time, size).
            torch.Tensor: Mask tensor (#batch, time, time).
            torch.Tensor: att_cache tensor,
                (#batch=1, head, cache_t1 + time, d_k * 2).
            torch.Tensor: cnn_cahce tensor (#batch, size, cache_t2).
        """

        # whether to use macaron style
        if self.feed_forward_macaron is not None:
            residual = x
            if self.normalize_before:
                x = self.norm_ff_macaron(x)
            x = residual + self.ff_scale * self.dropout(self.feed_forward_macaron(x))
            if not self.normalize_before:
                x = self.norm_ff_macaron(x)

        # multi-headed self-attention module
        residual = x
        if self.normalize_before:
            x = self.norm_mha(x)

        x_att, new_att_cache = self.self_attn(x, x, x, mask, pos_emb, att_cache)

        x = residual + self.dropout(x_att)
        if not self.normalize_before:
            x = self.norm_mha(x)

        # convolution module
        # Fake new cnn cache here, and then change it in conv_module
        new_cnn_cache = torch.tensor([0.0], dtype=x.dtype, device=x.device)
        if self.conv_module is not None:
            residual = x
            if self.normalize_before:
                x = self.norm_conv(x)
            x, new_cnn_cache = self.conv_module(x, mask_pad, cnn_cache)

            # add pointwise_conv for efficient conformer
            #   pointwise_conv_layer does not change shape
            if self.pointwise_conv_layer is not None:
                residual = residual.transpose(1, 2)
                residual = self.pointwise_conv_layer(residual)
                residual = residual.transpose(1, 2)
                assert residual.size(0) == x.size(0)
                assert residual.size(1) == x.size(1)
                assert residual.size(2) == x.size(2)

            x = residual + self.dropout(x)

            if not self.normalize_before:
                x = self.norm_conv(x)

        # feed forward module
        residual = x
        if self.normalize_before:
            x = self.norm_ff(x)

        x = residual + self.ff_scale * self.dropout(self.feed_forward(x))
        if not self.normalize_before:
            x = self.norm_ff(x)

        if self.conv_module is not None:
            x = self.norm_final(x)

        return x, mask, new_att_cache, new_cnn_cache