import torch
from torch import nn

from ..generic.normalization import LayerNorm


class DurationPredictor(nn.Module):
    """Glow-TTS duration prediction model.

    ::

        [2 x (conv1d_kxk -> relu -> layer_norm -> dropout)] -> conv1d_1x1 -> durs

    Args:
        in_channels (int): Number of channels of the input tensor.
        hidden_channels (int): Number of hidden channels of the network.
        kernel_size (int): Kernel size for the conv layers.
        dropout_p (float): Dropout rate used after each conv layer.
    """

    def __init__(self, in_channels, hidden_channels, kernel_size, dropout_p):
        super().__init__()
        # class arguments
        self.in_channels = in_channels
        self.filter_channels = hidden_channels
        self.kernel_size = kernel_size
        self.dropout_p = dropout_p
        # layers
        self.drop = nn.Dropout(dropout_p)
        self.conv_1 = nn.Conv1d(in_channels, hidden_channels, kernel_size, padding=kernel_size // 2)
        self.norm_1 = LayerNorm(hidden_channels)
        self.conv_2 = nn.Conv1d(hidden_channels, hidden_channels, kernel_size, padding=kernel_size // 2)
        self.norm_2 = LayerNorm(hidden_channels)
        # output layer
        self.proj = nn.Conv1d(hidden_channels, 1, 1)

    def forward(self, x, x_mask):
        """
        Shapes:
            - x: :math:`[B, C, T]`
            - x_mask: :math:`[B, 1, T]`
        """
        x = self.conv_1(x * x_mask)
        x = torch.relu(x)
        x = self.norm_1(x)
        x = self.drop(x)
        x = self.conv_2(x * x_mask)
        x = torch.relu(x)
        x = self.norm_2(x)
        x = self.drop(x)
        x = self.proj(x * x_mask)
        return x * x_mask