diff --git a/torchlpc/__init__.py b/torchlpc/__init__.py
index 620d360..6f3d1a6 100644
--- a/torchlpc/__init__.py
+++ b/torchlpc/__init__.py
@@ -1,5 +1,5 @@
 import torch
-from typing import Optional
+from typing import Optional, Union, Tuple
 from pathlib import Path
 import warnings
 
@@ -31,14 +31,18 @@
 
 
 def sample_wise_lpc(
-    x: torch.Tensor, a: torch.Tensor, zi: Optional[torch.Tensor] = None
-) -> torch.Tensor:
+    x: torch.Tensor,
+    a: torch.Tensor,
+    zi: Optional[torch.Tensor] = None,
+    return_zf: bool = False,
+) -> Union[torch.Tensor, Tuple[torch.Tensor, torch.Tensor]]:
     """Compute LPC filtering sample-wise.
 
     Args:
         x (torch.Tensor): Input signal.
         a (torch.Tensor): LPC coefficients.
         zi (torch.Tensor): Initial conditions.
+        return_zf (bool): If True, return the final filter delay values. Defaults to False.
 
     Shape:
         - x: :math:`(B, T)`
@@ -46,7 +50,8 @@ def sample_wise_lpc(
         - zi: :math:`(B, order)`
 
     Returns:
-        torch.Tensor: Filtered signal with the same shape as x.
+        Filtered signal with the same shape as x if `return_zf` is False.
+        If `return_zf` is True, returns a tuple of the filtered signal and the final delay values.
     """
     assert x.shape[0] == a.shape[0]
     assert x.shape[1] == a.shape[1]
@@ -62,6 +67,10 @@ def sample_wise_lpc(
     # if order == 1 and x.is_cuda and B * WARPSIZE < T:
     #     return RecurrenceCUDA.apply(-a.squeeze(2), x, zi.squeeze(1))
     if order == 1:
-        return Recurrence.apply(-a.squeeze(2), x, zi.squeeze(1))
+        y = Recurrence.apply(-a.squeeze(2), x, zi.squeeze(1))
+    else:
+        y = LPC.apply(x, a, zi)
 
-    return LPC.apply(x, a, zi)
+    if return_zf:
+        return y, y[:, -order:].flip(1)
+    return y