69 lines
2.2 KiB
Python
69 lines
2.2 KiB
Python
from typing import Dict, List
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
import torchaudio.transforms as T
|
|
|
|
|
|
class MultiResolutionSTFTLoss(nn.Module):
|
|
def __init__(
|
|
self,
|
|
fft_sizes: List[int] = [512, 1024, 2048, 4096, 8192],
|
|
hop_sizes: List[int] = [64, 128, 256, 512, 1024],
|
|
win_lengths: List[int] = [256, 512, 1024, 2048, 4096],
|
|
eps: float = 1e-7,
|
|
center: bool = True
|
|
):
|
|
super().__init__()
|
|
|
|
self.eps = eps
|
|
self.n_resolutions = len(fft_sizes)
|
|
|
|
self.stft_transforms = nn.ModuleList()
|
|
for i, (n_fft, hop_len, win_len) in enumerate(zip(fft_sizes, hop_sizes, win_lengths)):
|
|
stft = T.Spectrogram(
|
|
n_fft=n_fft,
|
|
hop_length=hop_len,
|
|
win_length=win_len,
|
|
window_fn=torch.hann_window,
|
|
power=None,
|
|
center=center,
|
|
pad_mode="reflect",
|
|
normalized=False,
|
|
)
|
|
self.stft_transforms.append(stft)
|
|
|
|
def forward(
|
|
self, y_true: torch.Tensor, y_pred: torch.Tensor
|
|
) -> Dict[str, torch.Tensor]:
|
|
if y_true.dim() == 3 and y_true.size(1) == 1:
|
|
y_true = y_true.squeeze(1)
|
|
if y_pred.dim() == 3 and y_pred.size(1) == 1:
|
|
y_pred = y_pred.squeeze(1)
|
|
|
|
sc_loss = 0.0
|
|
mag_loss = 0.0
|
|
|
|
for stft in self.stft_transforms:
|
|
stft.window = stft.window.to(y_true.device)
|
|
stft_true = stft(y_true)
|
|
stft_pred = stft(y_pred)
|
|
|
|
stft_mag_true = torch.abs(stft_true)
|
|
stft_mag_pred = torch.abs(stft_pred)
|
|
|
|
norm_true = torch.linalg.norm(stft_mag_true, dim=(-2, -1))
|
|
norm_diff = torch.linalg.norm(stft_mag_true - stft_mag_pred, dim=(-2, -1))
|
|
sc_loss += torch.mean(norm_diff / (norm_true + self.eps))
|
|
|
|
log_mag_pred = torch.log(stft_mag_pred + self.eps)
|
|
log_mag_true = torch.log(stft_mag_true + self.eps)
|
|
mag_loss += F.l1_loss(log_mag_pred, log_mag_true)
|
|
|
|
sc_loss /= self.n_resolutions
|
|
mag_loss /= self.n_resolutions
|
|
total_loss = sc_loss + mag_loss
|
|
|
|
return {"total": total_loss, "sc": sc_loss, "mag": mag_loss}
|