Merge pull request #36 from sp-nitech/update_mglsadf

Update mglsadf

.gitignore

@@ -1,6 +1,4 @@
 # project
-*.egg-info/
-build/
 dist/
 venv/
 
@@ -13,6 +11,7 @@ _build/
 
 # tools
 tools/SPTK/
+tools/toml/
 
 # misc
 __pycache__/

Makefile

@@ -22,23 +22,16 @@ init:
 	pip install -e .
 
 dev:
-	@if type virtualenv > /dev/null 2>&1; then \
-		test -d venv || virtualenv -p python$(PYTHON_VERSION) venv; \
-	else \
-		test -d venv || python$(PYTHON_VERSION) -m venv venv; \
-	fi
+	test -d venv || python$(PYTHON_VERSION) -m venv venv; \
 	. ./venv/bin/activate; pip install pip --upgrade; \
 	pip install torch==1.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html; \
 	pip install -e .[dev]
 
 dist:
-	./venv/bin/python setup.py bdist_wheel
-	./venv/bin/twine check dist/*
+	. ./venv/bin/activate; python -m build --wheel; \
+	twine check dist/*
 
 dist-clean:
-	@if [ -f ./venv/bin/python ]; then \
-		./venv/bin/python setup.py clean --all; \
-	fi
 	rm -rf dist
 
 doc:
@@ -72,21 +65,33 @@ test:
 	. ./venv/bin/activate; export PATH=tools/SPTK/bin:$$PATH; \
 		python -m pytest -s --cov=./ --cov-report=xml $$module
 
+test-clean:
+	rm -rf tests/__pycache__
+	rm -rf *.wav
+
 tool:
 	cd tools; make
 
 tool-clean:
 	cd tools; make clean
 
 update:
+	@if [ ! -x tools/toml/toml ]; then \
+		echo ""; \
+		echo "Error: please install toml-cli"; \
+		echo ""; \
+		echo "  make tool"; \
+		echo ""; \
+		exit 1; \
+	fi
 	./venv/bin/python -m pip install --upgrade pip
-	@for package in $$(cat setup.py | grep "           " | sed "s/\s//g" | \
-	sed 's/"//g' | tr ",\n" " "); do \
+	@for package in $$(./tools/toml/toml get pyproject.toml project.optional-dependencies.dev | \
+		sed 's/"//g' | tr -d '[]' | tr , ' '); do \
 		./venv/bin/pip install --upgrade $$package; \
 	done
 
-clean: dist-clean doc-clean tool-clean
-	rm -rf *.egg-info venv
+clean: dist-clean doc-clean test-clean tool-clean
+	rm -rf venv
 	find . -name "__pycache__" -type d | xargs rm -rf
 
-.PHONY: init dev dist dist-clean doc doc-clean check format test tool tool-clean update clean
+.PHONY: init dev dist dist-clean doc doc-clean check format test test-clean tool tool-clean update clean

README.md

@@ -62,7 +62,7 @@ mcep = diffsptk.MelCepstralAnalysis(cep_order=M, fft_length=n_fft, alpha=alpha,
 mc = mcep(X)
 
 # Reconstruct x.
-mlsa = diffsptk.MLSA(filter_order=M, alpha=alpha, frame_period=fp, taylor_order=30)
+mlsa = diffsptk.MLSA(filter_order=M, frame_period=fp, alpha=alpha, taylor_order=30)
 x_hat = mlsa(mlsa(x, -mc), mc)
 
 # Write reconstructed waveform.

diffsptk/core/dct.py

@@ -22,6 +22,19 @@
 
 
 def make_dct_matrix(L):
+    """Make DCT matrix.
+
+    Parameters
+    ----------
+    L : int >= 1 [scalar]
+        DCT length, :math:`L`.
+
+    Returns
+    -------
+    W : ndarray [shape=(L, L)]
+        DCT matrix.
+
+    """
     W = np.empty((L, L))
     n = (np.arange(L) + 0.5) * (np.pi / L)
     c = np.sqrt(2 / L)

diffsptk/core/delta.py

@@ -22,6 +22,23 @@
 
 
 def make_window(seed, static_out=True):
+    """Make window for delta computation.
+
+    Parameters
+    ----------
+    seed : list[list[float]] or list[int]
+        Delta coefficients or width(s) of 1st (and 2nd) regression coefficients.
+
+    static_out : bool [scalar]
+        If True, make window for static component.
+
+    Returns
+    -------
+    window : ndarray [shape=(D+1, W)]
+        Window for delta computation, where D is the number of dynamic components and
+        W is the width of window.
+
+    """
     if not isinstance(seed, (tuple, list)):
         raise ValueError("seed must be tuple or list")
 

diffsptk/core/fbank.py

@@ -26,8 +26,7 @@ def hz_to_mel(x):
 
 
 def sample_mel(n, fft_length, sample_rate):
-    hz = sample_rate * n / fft_length
-    return hz_to_mel(hz)
+    return hz_to_mel(sample_rate * n / fft_length)
 
 
 class MelFilterBankAnalysis(nn.Module):

diffsptk/core/freqt2.py

@@ -22,13 +22,33 @@
 
 
 def warp(omega, alpha, theta):
+    """Warp frequency.
+
+    Parameters
+    ----------
+    omega : float [0 <= omega <= 2pi]
+        Frequency.
+
+    alpha : float [-1 < alpha < 1]
+        Frequency warping factor, :math:`\\alpha`.
+
+    theta : float [0 <= theta <= 1]
+        Emphasis frequency, :math:`\\theta`.
+
+    Returns
+    -------
+    w : float
+        Warped frequency.
+
+    """
     x = omega - theta
     y = omega + theta
-    return (
+    w = (
         omega
         + np.arctan2(alpha * np.sin(x), 1 - alpha * np.cos(x))
         + np.arctan2(alpha * np.sin(y), 1 - alpha * np.cos(y))
     )
+    return w
 
 
 class SecondOrderAllPassFrequencyTransform(nn.Module):

diffsptk/core/imglsadf.py

@@ -20,53 +20,20 @@
 
 
 class PseudoInverseMGLSADigitalFilter(nn.Module):
-    """See `this page <https://sp-nitech.github.io/sptk/latest/main/imglsadf.html>`_
-    for details.
+    """See :func:`~diffsptk.PseudoMGLSADigitalFilter` for details."""
 
-    Parameters
-    ----------
-    filter_order : int >= 0 [scalar]
-        Order of filter coefficients, :math:`M`.
-
-    alpha : float [-1 < alpha < 1]
-        Frequency warping factor, :math:`\\alpha`.
-
-    gamma : float [-1 <= gamma <= 1]
-        Gamma, :math:`\\gamma`.
-
-    c : int >= 1 [scalar]
-        Number of stages.
-
-    frame_period : int >= 1 [scalar]
-        Frame period, :math:`P`.
-
-    ignore_gain : bool [scalar]
-        If True, perform filtering without gain.
-
-    phase : ['minimum', 'maximum', 'zero']
-        Filter type.
-
-    cascade : bool [scalar]
-        If True, use multi-stage FIR filter.
-
-    taylor_order : int >= 0 [scalar]
-        Order of Taylor series expansion (valid only if **cascade** is True).
-
-    ir_length : int >= 1 [scalar]
-        Length of impulse response (valid only if **cascade** is False).
-
-    n_fft : int >= 1 [scalar]
-        Number of FFT bins for conversion (valid only if **cascade** is False).
-
-    cep_order : int >= 0 [scalar]
-        Order of linear cepstrum (used to convert input to cepstrum).
-
-    """
-
-    def __init__(self, filter_order, **kwargs):
+    def __init__(self, filter_order, frame_period, **kwargs):
         super(PseudoInverseMGLSADigitalFilter, self).__init__()
 
-        self.mglsadf = PseudoMGLSADigitalFilter(filter_order, **kwargs)
+        # Change the default value of the order of Taylor series.
+        # This is because inverse filtering requires the large value.
+        if (
+            kwargs.get("mode", "multi-stage") == "multi-stage"
+            and "taylor_order" not in kwargs
+        ):
+            kwargs["taylor_order"] = 40
+
+        self.mglsadf = PseudoMGLSADigitalFilter(filter_order, frame_period, **kwargs)
 
     def forward(self, y, mc):
         """Apply an inverse MGLSA digital filter.

diffsptk/core/istft.py

@@ -22,7 +22,7 @@
 
 
 class InverseShortTermFourierTransform(nn.Module):
-    """This is the opposite module to ShortTermFourierTransform.
+    """This is the opposite module to :func:`~diffsptk.ShortTermFourierTransform`
 
     Parameters
     ----------

diffsptk/core/mgc2mgc.py

@@ -29,7 +29,28 @@
 
 
 class GeneralizedCepstrumToGeneralizedCepstrum(nn.Module):
-    def __init__(self, in_order, out_order, in_gamma, out_gamma, n_fft):
+    """Generalized cepstral transformation module.
+
+    Parameters
+    ----------
+    in_order : int >= 0 [scalar]
+        Order of input cepstrum, :math:`M_1`.
+
+    out_order : int >= 0 [scalar]
+        Order of output cepstrum, :math:`M_2`.
+
+    in_gamma : float [-1 <= in_gamma <= 1]
+        Input gamma, :math:`\\gamma_1`.
+
+    out_gamma : float [-1 <= out_gamma <= 1]
+        Output gamma, :math:`\\gamma_2`.
+
+    n_fft : int >> :math:`M_1, M_2` [scalar]
+        Number of FFT bins. Accurate conversion requires the large value.
+
+    """
+
+    def __init__(self, in_order, out_order, in_gamma, out_gamma, n_fft=512):
         super(GeneralizedCepstrumToGeneralizedCepstrum, self).__init__()
 
         self.in_order = in_order
@@ -45,6 +66,19 @@ def __init__(self, in_order, out_order, in_gamma, out_gamma, n_fft):
         assert max(self.in_order, self.out_order) + 1 < self.n_fft
 
     def forward(self, c1):
+        """Perform generalized cepstral transformation.
+
+        Parameters
+        ----------
+        c1 : Tensor [shape=(..., M1+1)]
+            Input cepstrum.
+
+        Returns
+        -------
+        c2 : Tensor [shape=(..., M2+1)]
+            Output cepstrum.
+
+        """
         c01 = torch.cat((c1[..., :1] * 0, c1[..., 1:]), dim=-1)
         C1 = torch.fft.fft(c01, n=self.n_fft)
 

diffsptk/core/mgc2sp.py

@@ -47,7 +47,7 @@ class MelGeneralizedCepstrumToSpectrum(nn.Module):
         If True, assume gamma-multiplied cepstrum.
 
     out_format : ['db', 'log-magnitude', 'magnitude', 'power', \
-                  'cycle', 'radian', 'degree']
+                  'cycle', 'radian', 'degree', 'complex']
         Output format.
 
     n_fft : int >> :math:`L` [scalar]
@@ -88,6 +88,8 @@ def __init__(
         elif out_format == 6 or out_format == "degree":
             c = 180 / math.pi
             self.convert = lambda x: x.imag * c
+        elif out_format == "complex":
+            self.convert = lambda x: torch.polar(torch.exp(x.real), x.imag)
         else:
             raise ValueError(f"out_format {out_format} is not supported")