NeuralSeq/modules/commons/normalizing_flow/utils.py

import torch


def squeeze(x, x_mask=None, n_sqz=2):
    b, c, t = x.size()

    t = (t // n_sqz) * n_sqz
    x = x[:, :, :t]
    x_sqz = x.view(b, c, t // n_sqz, n_sqz)
    x_sqz = x_sqz.permute(0, 3, 1, 2).contiguous().view(b, c * n_sqz, t // n_sqz)

    if x_mask is not None:
        x_mask = x_mask[:, :, n_sqz - 1::n_sqz]
    else:
        x_mask = torch.ones(b, 1, t // n_sqz).to(device=x.device, dtype=x.dtype)
    return x_sqz * x_mask, x_mask


def unsqueeze(x, x_mask=None, n_sqz=2):
    b, c, t = x.size()

    x_unsqz = x.view(b, n_sqz, c // n_sqz, t)
    x_unsqz = x_unsqz.permute(0, 2, 3, 1).contiguous().view(b, c // n_sqz, t * n_sqz)

    if x_mask is not None:
        x_mask = x_mask.unsqueeze(-1).repeat(1, 1, 1, n_sqz).view(b, 1, t * n_sqz)
    else:
        x_mask = torch.ones(b, 1, t * n_sqz).to(device=x.device, dtype=x.dtype)
    return x_unsqz * x_mask, x_mask