Merge pull request #101 from sp-nitech/pca

Support batch input in pca

Makefile

@@ -51,7 +51,7 @@ check: tool
 	. ./venv/bin/activate && python -m ruff format --check $(PROJECT) tests docs
 	. ./venv/bin/activate && python -m isort --check $(PROJECT) tests
 	. ./venv/bin/activate && python -m mdformat --check *.md
-	. ./venv/bin/activate && python -m docstrfmt --check docs
+	. ./venv/bin/activate && cd docs && python -m docstrfmt --check .
 	./tools/taplo/taplo fmt --check *.toml
 	./tools/yamlfmt/yamlfmt --lint *.cff *.yml .github/workflows/*.yml
 
@@ -60,7 +60,7 @@ format: tool
 	. ./venv/bin/activate && python -m ruff format $(PROJECT) tests docs
 	. ./venv/bin/activate && python -m isort $(PROJECT) tests
 	. ./venv/bin/activate && python -m mdformat *.md
-	. ./venv/bin/activate && python -m docstrfmt docs
+	. ./venv/bin/activate && cd docs && python -m docstrfmt --check .
 	./tools/taplo/taplo fmt *.toml
 	./tools/yamlfmt/yamlfmt *.cff *.yml .github/workflows/*.yml
 

diffsptk/functional.py

@@ -175,7 +175,7 @@ def c2ndps(c, fft_length):
     )
 
 
-def cdist(c1, c2, full=False, reduction="mean", eps=1e-8):
+def cdist(c1, c2, full=False, reduction="mean"):
     """Calculate cepstral distance between two inputs.
 
     Parameters
@@ -192,16 +192,13 @@ def cdist(c1, c2, full=False, reduction="mean", eps=1e-8):
     reduction : ['none', 'mean', 'batchmean', 'sum']
         Reduction type.
 
-    eps : float >= 0
-        A small value to prevent NaN.
-
     Returns
     -------
     out : Tensor [shape=(...,) or scalar]
         Cepstral distance.
 
     """
-    return nn.CepstralDistance._func(c1, c2, full=full, reduction=reduction, eps=eps)
+    return nn.CepstralDistance._func(c1, c2, full=full, reduction=reduction)
 
 
 def chroma(x, n_channel, sample_rate, norm=float("inf")):
@@ -1953,7 +1950,7 @@ def rlevdur(a):
     return nn.ReverseLevinsonDurbin._func(a)
 
 
-def rmse(x, y, reduction="mean", eps=1e-8):
+def rmse(x, y, reduction="mean"):
     """Calculate RMSE.
 
     Parameters
@@ -1967,16 +1964,13 @@ def rmse(x, y, reduction="mean", eps=1e-8):
     reduction : ['none', 'mean', 'sum']
         Reduction type.
 
-    eps : float >= 0
-        A small value to prevent NaN.
-
     Returns
     -------
     out : Tensor [shape=(...,) or scalar]
         RMSE.
 
     """
-    return nn.RootMeanSquareError._func(x, y, reduction=reduction, eps=eps)
+    return nn.RootMeanSquareError._func(x, y, reduction=reduction)
 
 
 def root_pol(a, out_format="rectangular"):
@@ -2074,88 +2068,6 @@ def spec(
     )
 
 
-def ssim(
-    x,
-    y,
-    reduction="mean",
-    *,
-    alpha=1,
-    beta=1,
-    gamma=1,
-    kernel_size=11,
-    sigma=1.5,
-    k1=0.01,
-    k2=0.03,
-    eps=1e-8,
-    padding="same",
-    dynamic_range=None,
-):
-    """Calculate SSIM.
-
-    Parameters
-    ----------
-    x : Tensor [shape=(..., N, D)]
-        Input.
-
-    y : Tensor [shape=(..., N, D)]
-        Target.
-
-    reduction : ['none', 'mean', 'sum']
-        Reduction type.
-
-    alpha : float > 0
-        Relative importance of luminance component.
-
-    beta : float > 0
-        Relative importance of contrast component.
-
-    gamma : float > 0
-        Relative importance of structure component.
-
-    kernel_size : int >= 1
-        Kernel size of Gaussian filter.
-
-    sigma : float > 0
-        Standard deviation of Gaussian filter.
-
-    k1 : float > 0
-        A small constant.
-
-    k2 : float > 0
-        A small constant.
-
-    eps : float >= 0
-        A small value to prevent NaN.
-
-    padding : ['valid', 'same']
-        Padding type.
-
-    dynamic_range : float > 0 or None
-        Dynamic range of input. If None, input is automatically normalized.
-
-    Returns
-    -------
-    out : Tensor [shape=(..., N, D) or scalar]
-        SSIM or mean SSIM.
-
-    """
-    return nn.StructuralSimilarityIndex._func(
-        x,
-        y,
-        reduction=reduction,
-        alpha=alpha,
-        beta=beta,
-        gamma=gamma,
-        kernel_size=kernel_size,
-        sigma=sigma,
-        k1=k1,
-        k2=k2,
-        eps=eps,
-        padding=padding,
-        dynamic_range=dynamic_range,
-    )
-
-
 def stft(
     x,
     *,

diffsptk/misc/utils.py

@@ -254,6 +254,12 @@ def clog(x):
     return torch.log(x.abs())
 
 
+def outer(x, y=None):
+    return torch.matmul(
+        x.unsqueeze(-1), x.unsqueeze(-2) if y is None else y.unsqueeze(-2)
+    )
+
+
 def iir(x, b=None, a=None):
     if b is None:
         b = torch.ones(1, dtype=x.dtype, device=x.device)

diffsptk/modules/__init__.py

@@ -124,8 +124,6 @@
 from .snr import SignalToNoiseRatio
 from .snr import SignalToNoiseRatio as SNR
 from .spec import Spectrum
-from .ssim import StructuralSimilarityIndex
-from .ssim import StructuralSimilarityIndex as SSIM
 from .stft import ShortTimeFourierTransform
 from .stft import ShortTimeFourierTransform as STFT
 from .ulaw import MuLawCompression

diffsptk/modules/cdist.py

@@ -30,20 +30,15 @@ class CepstralDistance(nn.Module):
     reduction : ['none', 'mean', 'batchmean', 'sum']
         Reduction type.
 
-    eps : float >= 0
-        A small value to prevent NaN.
-
     """
 
-    def __init__(self, full=False, reduction="mean", eps=1e-8):
+    def __init__(self, full=False, reduction="mean"):
         super().__init__()
 
         assert reduction in ("none", "mean", "batchmean", "sum")
-        assert 0 <= eps
 
         self.full = full
         self.reduction = reduction
-        self.eps = eps
 
     def forward(self, c1, c2):
         """Calculate cepstral distance between two inputs.
@@ -75,11 +70,11 @@ def forward(self, c1, c2):
         tensor(1.6551)
 
         """
-        return self._forward(c1, c2, self.full, self.reduction, self.eps)
+        return self._forward(c1, c2, self.full, self.reduction)
 
     @staticmethod
-    def _forward(c1, c2, full, reduction, eps):
-        distance = torch.sqrt((c1[..., 1:] - c2[..., 1:]).square().sum(-1) + eps)
+    def _forward(c1, c2, full, reduction):
+        distance = torch.linalg.vector_norm(c1[..., 1:] - c2[..., 1:], ord=2, dim=-1)
 
         if reduction == "none":
             pass

diffsptk/modules/gmm.py

@@ -21,12 +21,13 @@
 from torch import nn
 from tqdm import tqdm
 
+from ..misc.utils import outer
 from ..misc.utils import to_dataloader
 
 
 class GaussianMixtureModeling(nn.Module):
     """See `this page <https://sp-nitech.github.io/sptk/latest/main/gmm.html>`_
-    for details. This module is not differentiable.
+    for details. Note that the forward method is not differentiable.
 
     Parameters
     ----------
@@ -210,11 +211,10 @@ def warmup(self, x, **lbg_params):
         b = 0
         for (batch_x,) in tqdm(x, disable=self.hide_progress_bar):
             e = b + batch_x.size(0)
-            xp = batch_x.to(device)
-            xx = torch.matmul(xp.unsqueeze(-1), xp.unsqueeze(-2))
+            xx = outer(batch_x.to(device))
             kxx.scatter_add_(0, idx[b:e], xx)
             b = e
-        mm = torch.matmul(mu.unsqueeze(-1), mu.unsqueeze(-2))  # (K, L, L)
+        mm = outer(mu)  # (K, L, L)
         sigma = kxx / count.view(-1, 1, 1) - mm
         sigma = sigma * self.mask
 
@@ -328,24 +328,21 @@ def forward(self, x, return_posterior=False):
                 b = 0
                 for (batch_x,) in tqdm(x, disable=self.hide_progress_bar):
                     e = b + batch_x.size(0)
-                    xp = batch_x.to(device)
-                    xx = torch.matmul(xp.unsqueeze(-1), xp.unsqueeze(-2))
+                    xx = outer(batch_x.to(device))
                     pxx.append(torch.einsum("bk,blm->klm", posterior[b:e], xx))
                     b = e
                 pxx = sum(pxx)
-                mm = torch.matmul(self.mu.unsqueeze(-1), self.mu.unsqueeze(-2))
+                mm = outer(self.mu)
                 if self.alpha == 0:
                     sigma = pxx * z.view(-1, 1, 1) - mm
                 else:
                     y = posterior.sum(dim=0)
                     nu = px / y.view(-1, 1)
-                    nm = torch.matmul(nu.unsqueeze(-1), self.mu.unsqueeze(-2))
+                    nm = outer(nu, self.mu)
                     mn = nm.transpose(-2, -1)
                     a = pxx - y.view(-1, 1, 1) * (nm + mn - mm)
                     b = xi.view(-1, 1, 1) * self.ubm_sigma
-                    diff = self.ubm_mu - self.mu
-                    dd = torch.matmul(diff.unsqueeze(-1), diff.unsqueeze(-2))
-                    c = xi.view(-1, 1, 1) * dd
+                    c = xi.view(-1, 1, 1) * outer(self.ubm_mu - self.mu)
                     sigma = (a + b + c) * z.view(-1, 1, 1)
                 self.sigma = sigma * self.mask
             self.sigma.diagonal(dim1=-2, dim2=-1).clamp_(min=self.var_floor)

diffsptk/modules/lbg.py

@@ -29,7 +29,7 @@
 
 class LindeBuzoGrayAlgorithm(nn.Module):
     """See `this page <https://sp-nitech.github.io/sptk/latest/main/lbg.html>`_
-    for details. This module is not differentiable.
+    for details. Note that the forward method is not differentiable.
 
     Parameters
     ----------