tts/modules/wavvae/encoder/common_modules/conv.py

# MIT License

# Copyright (c) Meta Platforms, Inc. and affiliates.

# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:

# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.

# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

# Copyright (c) [2023] [Meta Platforms, Inc. and affiliates.] 
# Copyright (c) [2025] [Ziyue Jiang] 
# SPDX-License-Identifier: MIT
# This file has been modified by Ziyue Jiang on 2025/03/19
# Original file was released under MIT, with the full license text # available at https://github.com/facebookresearch/encodec/blob/gh-pages/LICENSE.
# This modified file is released under the same license.

"""Convolutional layers wrappers and utilities."""

import math
import typing as tp
import warnings
import einops

import torch
from torch import nn
from torch.nn import functional as F
from torch.nn.utils import spectral_norm, weight_norm


CONV_NORMALIZATIONS = frozenset(['none', 'weight_norm', 'spectral_norm',
                                 'time_layer_norm', 'layer_norm', 'time_group_norm'])


def apply_parametrization_norm(module: nn.Module, norm: str = 'none') -> nn.Module:
    assert norm in CONV_NORMALIZATIONS
    if norm == 'weight_norm':
        return weight_norm(module)
    elif norm == 'spectral_norm':
        return spectral_norm(module)
    else:
        return module


def get_norm_module(module: nn.Module, causal: bool = False, norm: str = 'none', **norm_kwargs) -> nn.Module:
    assert norm in CONV_NORMALIZATIONS
    if norm == 'layer_norm':
        assert isinstance(module, nn.modules.conv._ConvNd)
        return ConvLayerNorm(module.out_channels, **norm_kwargs)
    elif norm == 'time_group_norm':
        if causal:
            raise ValueError("GroupNorm doesn't support causal evaluation.")
        assert isinstance(module, nn.modules.conv._ConvNd)
        return nn.GroupNorm(1, module.out_channels, **norm_kwargs)
    else:
        return nn.Identity()


def get_extra_padding_for_conv1d(x: torch.Tensor, kernel_size: int, stride: int,
                                 padding_total: int = 0) -> int:
    length = x.shape[-1]
    n_frames = (length - kernel_size + padding_total) / stride + 1
    ideal_length = (math.ceil(n_frames) - 1) * stride + (kernel_size - padding_total)
    return ideal_length - length


def pad1d(x: torch.Tensor, paddings: tp.Tuple[int, int], mode: str = 'zero', value: float = 0.):
    length = x.shape[-1]
    padding_left, padding_right = paddings
    assert padding_left >= 0 and padding_right >= 0, (padding_left, padding_right)
    if mode == 'reflect':
        max_pad = max(padding_left, padding_right)
        extra_pad = 0
        if length <= max_pad:
            extra_pad = max_pad - length + 1
            x = F.pad(x, (0, extra_pad))
        padded = F.pad(x, paddings, mode, value)
        end = padded.shape[-1] - extra_pad
        return padded[..., :end]
    else:
        return F.pad(x, paddings, mode, value)


class ConvLayerNorm(nn.LayerNorm):
    def __init__(self, normalized_shape: tp.Union[int, tp.List[int], torch.Size], **kwargs):
        super().__init__(normalized_shape, **kwargs)

    def forward(self, x):
        x = einops.rearrange(x, 'b ... t -> b t ...')
        x = super().forward(x)
        x = einops.rearrange(x, 'b t ... -> b ... t')
        return


class NormConv1d(nn.Module):
    def __init__(self, *args, causal: bool = False, norm: str = 'none',
                 norm_kwargs: tp.Dict[str, tp.Any] = {}, **kwargs):
        super().__init__()
        self.conv = apply_parametrization_norm(nn.Conv1d(*args, **kwargs), norm)
        self.norm = get_norm_module(self.conv, causal, norm, **norm_kwargs)
        self.norm_type = norm

    def forward(self, x):
        x = self.conv(x)
        x = self.norm(x)
        return x


class SConv1d(nn.Module):
    def __init__(self, in_channels: int, out_channels: int,
                 kernel_size: int, stride: int = 1, dilation: int = 1,
                 groups: int = 1, bias: bool = True, causal: bool = False,
                 norm: str = 'none', norm_kwargs: tp.Dict[str, tp.Any] = {},
                 pad_mode: str = 'reflect'):
        super().__init__()
        # warn user on unusual setup between dilation and stride
        if stride > 1 and dilation > 1:
            warnings.warn('SConv1d has been initialized with stride > 1 and dilation > 1'
                          f' (kernel_size={kernel_size} stride={stride}, dilation={dilation}).')
        self.conv = NormConv1d(in_channels, out_channels, kernel_size, stride,
                               dilation=dilation, groups=groups, bias=bias, causal=causal,
                               norm=norm, norm_kwargs=norm_kwargs)
        self.causal = causal
        self.pad_mode = pad_mode

    def forward(self, x):
        B, C, T = x.shape
        kernel_size = self.conv.conv.kernel_size[0]
        stride = self.conv.conv.stride[0]
        dilation = self.conv.conv.dilation[0]
        kernel_size = (kernel_size - 1) * dilation + 1  # effective kernel size with dilations
        padding_total = kernel_size - stride
        extra_padding = get_extra_padding_for_conv1d(x, kernel_size, stride, padding_total)
        if self.causal:
            # Left padding for causal
            x = pad1d(x, (padding_total, extra_padding), mode=self.pad_mode)
        else:
            # Asymmetric padding required for odd strides
            padding_right = padding_total // 2
            padding_left = padding_total - padding_right
            x = pad1d(x, (padding_left, padding_right + extra_padding), mode=self.pad_mode)
        return self.conv(x)