Merge pull request #143 from sp-nitech/griffin

Maintenance

Author: takenori-y

diffsptk/modules/drc.py

@@ -16,10 +16,9 @@
 
 import numpy as np
 import torch
-import torchcomp
 from torch import nn
 
-from ..typing import Precomputed
+from ..typing import Callable, Precomputed
 from ..utils.private import filter_values, to, to_2d
 from .base import BaseFunctionalModule
 
@@ -161,6 +160,8 @@ def _precompute(
         device: torch.device | None,
         dtype: torch.dtype | None,
     ) -> Precomputed:
+        import torchcomp
+
         DynamicRangeCompression._check(
             ratio, attack_time, release_time, sample_rate, makeup_gain, abs_max
         )
@@ -179,20 +180,21 @@ def _precompute(
             makeup_gain = to(torch.tensor(makeup_gain, device=device), dtype=dtype)
         makeup_gain = 10 ** (makeup_gain / 20)
         params = torch.stack([threshold, ratio, attack_time, release_time, makeup_gain])
-        return (abs_max,), None, (params,)
+        return (abs_max, torchcomp.compexp_gain), None, (params,)
 
     @staticmethod
     def _forward(
         x: torch.Tensor,
         abs_max: float,
+        compexp_gain: Callable,
         params: torch.Tensor,
     ) -> torch.Tensor:
         eps = 1e-10
 
         y = to_2d(x)
         y_abs = y.abs() / abs_max + eps
 
-        g = torchcomp.compexp_gain(
+        g = compexp_gain(
             y_abs,
             params[0],
             params[1],

diffsptk/modules/griffin.py

@@ -54,7 +54,7 @@ class GriffinLim(BaseFunctionalModule):
     symmetric : bool
         If True, the window is symmetric, otherwise periodic.
 
-    n_iter : int >= 1
+    n_iter : int >= 0
         The number of iterations for phase reconstruction.
 
     alpha : float >= 0
@@ -159,8 +159,8 @@ def _check(
         beta: float,
         gamma: float,
     ) -> None:
-        if n_iter <= 0:
-            raise ValueError("n_iter must be positive.")
+        if n_iter < 0:
+            raise ValueError("n_iter must be non-negative.")
         if alpha < 0:
             raise ValueError("alpha must be non-negative.")
         if beta < 0:
@@ -259,7 +259,8 @@ def _forward(
         if logger is not None:
             logger.info(f"alpha: {alpha}, beta: {beta}, gamma: {gamma}")
 
-        s = torch.sqrt(y)
+        eps = 1e-16
+        s = torch.sqrt(y + eps)
         angle = torch.exp(1j * phase_generator(s))
 
         t_prev = d_prev = 0  # This suppresses F821 and F841.
@@ -275,7 +276,7 @@ def _forward(
                 c = t + alpha * diff
                 d = t + beta * diff
 
-            angle = c / (c.abs() + 1e-16)
+            angle = c / (c.abs() + eps)
             t_prev = t
             d_prev = d
 

diffsptk/modules/poledf.py

@@ -15,9 +15,8 @@
 # ------------------------------------------------------------------------ #
 
 import torch
-from torchlpc import sample_wise_lpc
 
-from ..typing import Precomputed
+from ..typing import Callable, Precomputed
 from ..utils.private import check_size, filter_values
 from .base import BaseFunctionalModule
 from .linear_intpl import LinearInterpolation
@@ -105,11 +104,17 @@ def _precompute(
         filter_order: int, frame_period: int, ignore_gain: bool = False
     ) -> Precomputed:
         AllPoleDigitalFilter._check(filter_order, frame_period)
-        return (frame_period, ignore_gain)
+        from torchlpc import sample_wise_lpc
+
+        return (frame_period, ignore_gain, sample_wise_lpc)
 
     @staticmethod
     def _forward(
-        x: torch.Tensor, a: torch.Tensor, frame_period: int, ignore_gain: bool
+        x: torch.Tensor,
+        a: torch.Tensor,
+        frame_period: int,
+        ignore_gain: bool,
+        sample_wise_lpc: Callable,
     ) -> torch.Tensor:
         check_size(x.size(-1), a.size(-2) * frame_period, "sequence length")
 

diffsptk/utils/public.py

@@ -190,5 +190,5 @@ def write(
     >>> diffsptk.write("out.wav", x, sr)
 
     """
-    x = x.cpu().numpy() if torch.is_tensor(x) else x
+    x = x.detach().cpu().numpy() if torch.is_tensor(x) else x
     sf.write(filename, x.T if channel_first else x, sr, **kwargs)