diff --git a/README.md b/README.md
index c7eed13..d5702d1 100644
--- a/README.md
+++ b/README.md
@@ -142,4 +142,4 @@ You can install dependencies these by `pip install -r requirements.txt`
 
 
 ## Acknowledgement
-Thank all contributors ([tats-u](https://github.com/tats-u), [wuaalb](https://github.com/wuaalb), [r9y9](https://github.com/r9y9), [rikrd](https://github.com/rikrd), [kudan2510](https://github.com/kundan2510)) for making this repo better and [sotelo](https://github.com/sotelo) whose [world.py](https://github.com/sotelo/world.py) inspired this repo.<br/>
+Thank all contributors ([tats-u](https://github.com/tats-u), [wuaalb](https://github.com/wuaalb), [r9y9](https://github.com/r9y9), [rikrd](https://github.com/rikrd), [kudan2510](https://github.com/kundan2510), [sabonerune](https://github.com/sabonerune)) for making this repo better and [sotelo](https://github.com/sotelo) whose [world.py](https://github.com/sotelo/world.py) inspired this repo.<br/>
diff --git a/appveyor.yml b/appveyor.yml
index 0fa3ee4..b71e0ab 100644
--- a/appveyor.yml
+++ b/appveyor.yml
@@ -1,3 +1,7 @@
+# Do not build a feature branch that already has an open pull request
+# (otherwise every PR update triggers two identical builds).
+skip_branch_with_pr: true
+
 environment:
   TWINE_USERNAME: __token__
   TWINE_PASSWORD:
@@ -48,6 +52,26 @@ environment:
       PYTHON_ARCH: "64"
       MINICONDA: C:\Miniconda3-x64
 
+    # - PYTHON_VERSION: "3.11"
+    #   PYTHON_ARCH: "32"
+    #   MINICONDA: C:\Miniconda3
+
+    - PYTHON_VERSION: "3.12"
+      PYTHON_ARCH: "64"
+      MINICONDA: C:\Miniconda3-x64
+
+    # - PYTHON_VERSION: "3.12"
+    #   PYTHON_ARCH: "32"
+    #   MINICONDA: C:\Miniconda3
+
+    - PYTHON_VERSION: "3.13"
+      PYTHON_ARCH: "64"
+      MINICONDA: C:\Miniconda3-x64
+
+    # - PYTHON_VERSION: "3.13"
+    #   PYTHON_ARCH: "32"
+    #   MINICONDA: C:\Miniconda3
+
 # The above commented lines has problem executing the `test_script`
 # Python >= 3.8 (32-bit)
 
@@ -70,7 +94,9 @@ install:
   - conda update -q conda
   - conda config --add channels pypi
   - conda info -a
-  - "conda create -q -n test-environment python=%PYTHON_VERSION% numpy cython wheel matplotlib"
+  # matplotlib/Pillow are installed from PyPI in `test_script`: conda's win-64
+  # Pillow for py3.6/3.7 ships a `_imaging` DLL that fails to load.
+  - "conda create -q -n test-environment python=%PYTHON_VERSION% numpy cython wheel"
   - activate test-environment
   - python -m pip install --no-cache-dir -U pip
   - ps: |
@@ -82,10 +108,12 @@ build_script:
   - pip install -e .
 
 test_script:
+  - pip install "soundfile<0.13; python_version < '3.8'" "soundfile; python_version >= '3.8'"
+  - "pip install --only-binary=:all: matplotlib"
   - ps: |
-      $ErrorActionPreference = "Stop"
       Set-Location "$env:APPVEYOR_BUILD_FOLDER\demo"
       python demo.py
+      if ($LASTEXITCODE -ne 0) { exit $LASTEXITCODE }
       Set-Location "$env:APPVEYOR_BUILD_FOLDER"
 
 after_test:
diff --git a/lib/World b/lib/World
index 8d79b88..d625e76 160000
--- a/lib/World
+++ b/lib/World
@@ -1 +1 @@
-Subproject commit 8d79b88b7dd92e8a132996cf74080b2d6f881b98
+Subproject commit d625e7608ca23a870018f01e7c562ac683d9847f
diff --git a/pyproject.toml b/pyproject.toml
index b8eaa23..380d95b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,5 +4,6 @@ requires = [
     "numpy>=2; python_version>='3.9'",
     "oldest-supported-numpy; python_version<'3.9'",
     "cython",
+    "cython<3.1.0; python_version<'3.9'",
 ]
 build-backend = "setuptools.build_meta"
diff --git a/pyworld/__init__.py b/pyworld/__init__.py
index 73ba9e3..5e061cd 100644
--- a/pyworld/__init__.py
+++ b/pyworld/__init__.py
@@ -10,8 +10,16 @@
 
 from __future__ import division, print_function, absolute_import
 
-import pkg_resources
+import sys
 
-__version__ = pkg_resources.get_distribution('pyworld').version
+if sys.version_info >= (3, 8):
+    from importlib.metadata import version, PackageNotFoundError
+else:
+    from importlib_metadata import version, PackageNotFoundError
+
+try:
+    __version__ = version('pyworld')
+except PackageNotFoundError:
+    __version__ = '0.3.6'
 
 from .pyworld import *
diff --git a/pyworld/py.typed b/pyworld/py.typed
new file mode 100644
index 0000000..17af580
--- /dev/null
+++ b/pyworld/py.typed
@@ -0,0 +1 @@
+# PEP 561
diff --git a/pyworld/pyworld.pyi b/pyworld/pyworld.pyi
new file mode 100644
index 0000000..49e9069
--- /dev/null
+++ b/pyworld/pyworld.pyi
@@ -0,0 +1,396 @@
+from typing import Optional, Tuple
+
+import numpy as np
+
+default_frame_period: float
+default_f0_floor: float
+default_f0_ceil: float
+
+def dio(
+    x: np.ndarray,
+    fs: int,
+    f0_floor: float = ...,
+    f0_ceil: float = ...,
+    channels_in_octave: float = ...,
+    frame_period: float = ...,
+    speed: int = ...,
+    allowed_range: float = ...,
+) -> Tuple[np.ndarray, np.ndarray]:
+    """DIO F0 extraction algorithm.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    fs : int
+        Sample rate of input signal in Hz.
+    f0_floor : float
+        Lower F0 limit in Hz.
+        Default: 71.0
+    f0_ceil : float
+        Upper F0 limit in Hz.
+        Default: 800.0
+    channels_in_octave : float
+        Resolution of multiband processing; normally shouldn't be changed.
+        Default: 2.0
+    frame_period : float
+        Period between consecutive frames in milliseconds.
+        Default: 5.0
+    speed : int
+        The F0 estimator may downsample the input signal using this integer factor
+        (range [1;12]). The algorithm will then operate on a signal at fs/speed Hz
+        to reduce computational complexity, but high values may negatively impact
+        accuracy.
+        Default: 1 (no downsampling)
+    allowed_range : float
+        Threshold for voiced/unvoiced decision. Can be any value >= 0, but 0.02 to 0.2
+        is a reasonable range. Lower values will cause more frames to be considered
+        unvoiced (in the extreme case of `threshold=0`, almost all frames will be unvoiced).
+        Default: 0.1
+
+    Returns
+    -------
+    f0 : ndarray
+        Estimated F0 contour.
+    temporal_positions : ndarray
+        Temporal position of each frame.
+    """
+    ...
+
+def harvest(
+    x: np.ndarray,
+    fs: int,
+    f0_floor: float = ...,
+    f0_ceil: float = ...,
+    frame_period: float = ...,
+) -> Tuple[np.ndarray, np.ndarray]:
+    """Harvest F0 extraction algorithm.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    fs : int
+        Sample rate of input signal in Hz.
+    f0_floor : float
+        Lower F0 limit in Hz.
+        Default: 71.0
+    f0_ceil : float
+        Upper F0 limit in Hz.
+        Default: 800.0
+    frame_period : float
+        Period between consecutive frames in milliseconds.
+        Default: 5.0
+
+    Returns
+    -------
+    f0 : ndarray
+        Estimated F0 contour.
+    temporal_positions : ndarray
+        Temporal position of each frame.
+    """
+    ...
+
+def stonemask(
+    x: np.ndarray, f0: np.ndarray, temporal_positions: np.ndarray, fs: int
+) -> np.ndarray:
+    """StoneMask F0 refinement algorithm.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    f0 : ndarray
+        Input F0 contour.
+    temporal_positions : ndarray
+        Temporal positions of each frame.
+    fs : int
+        Sample rate of input signal in Hz.
+
+    Returns
+    -------
+    refined_f0 : ndarray
+        Refined F0 contour.
+    """
+    ...
+
+def get_cheaptrick_fft_size(fs: int, f0_floor: float = ...) -> int:
+    """Calculate suitable FFT size for CheapTrick given F0 floor.
+
+    Parameters
+    ----------
+    fs : int
+        Sample rate of input signal in Hz.
+    f0_floor : float
+        Lower F0 limit in Hz. The required FFT size is a direct
+        consequence of the F0 floor used.
+        Default: 71.0
+
+    Returns
+    -------
+    fft_size : int
+        Resulting FFT size.
+    """
+    ...
+
+def get_cheaptrick_f0_floor(fs: int, fft_size: int) -> float:
+    """Calculates actual lower F0 limit for CheapTrick
+    based on the sampling frequency and FFT size used. Whenever F0 is below
+    this threshold the spectrum will be analyzed as if the frame is unvoiced
+    (using kDefaultF0 defined in constantnumbers.h).
+
+    Parameters
+    ----------
+    fs : int
+        Sample rate of input signal in Hz.
+    fft_size : int
+        FFT size used for CheapTrick.
+
+    Returns
+    -------
+    f0_floor : float
+        Resulting lower F0 limit in Hz.
+    """
+    ...
+
+def cheaptrick(
+    x: np.ndarray,
+    f0: np.ndarray,
+    temporal_positions: np.ndarray,
+    fs: int,
+    q1: float = ...,
+    f0_floor: float = ...,
+    fft_size: Optional[int] = ...,
+) -> np.ndarray:
+    """CheapTrick harmonic spectral envelope estimation algorithm.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    f0 : ndarray
+        Input F0 contour.
+    temporal_positions : ndarray
+        Temporal positions of each frame.
+    fs : int
+        Sample rate of input signal in Hz.
+    q1 : float
+        Spectral recovery parameter.
+        Default: -0.15 (this value was tuned and normally does not need adjustment)
+    f0_floor : float, None
+        Lower F0 limit in Hz. Not used in case `fft_size` is specified.
+        Default: 71.0
+    fft_size : int, None
+        FFT size to be used. When `None` (default) is used, the FFT size is computed
+        automatically as a function of the given input sample rate and F0 floor.
+        When `fft_size` is specified, the given `f0_floor` parameter is ignored.
+        Default: None
+
+    Returns
+    -------
+    spectrogram : ndarray
+        Spectral envelope (squared magnitude).
+    """
+    ...
+
+def d4c(
+    x: np.ndarray,
+    f0: np.ndarray,
+    temporal_positions: np.ndarray,
+    fs: int,
+    threshold: float = ...,
+    fft_size: Optional[int] = ...,
+) -> np.ndarray:
+    """D4C aperiodicity estimation algorithm.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    f0 : ndarray
+        Input F0 contour.
+    temporal_positions : ndarray
+        Temporal positions of each frame.
+    fs : int
+        Sample rate of input signal in Hz.
+    threshold : float
+        Threshold for aperiodicity-based voiced/unvoiced decision, in range 0 to 1.
+        If a value of 0 is used, voiced frames will be kept voiced. If a value > 0 is
+        used some voiced frames can be considered unvoiced by setting their aperiodicity
+        to 1 (thus synthesizing them with white noise). Using `threshold=0` will result
+        in the behavior of older versions of D4C. The current default of 0.85 is meant
+        to be used in combination with the Harvest F0 estimator, which was designed to have
+        a high voiced/unvoiced threshold (i.e. most frames will be considered voiced).
+        Default: 0.85
+    fft_size : int, None
+        FFT size to be used. When `None` (default) is used, the FFT size is computed
+        automatically as a function of the given input sample rate and the default F0 floor.
+        When `fft_size` is specified, it should match the FFT size used to compute
+        the spectral envelope (i.e. `fft_size=2*(sp.shape[1] - 1)`) in order to get the
+        desired results when resynthesizing.
+        Default: None
+
+    Returns
+    -------
+    aperiodicity : ndarray
+        Aperiodicity (envelope, linear magnitude relative to spectral envelope).
+    """
+    ...
+
+def synthesize(
+    f0: np.ndarray,
+    spectrogram: np.ndarray,
+    aperiodicity: np.ndarray,
+    fs: int,
+    frame_period: float = ...,
+) -> np.ndarray:
+    """WORLD synthesis from parametric representation.
+
+    Parameters
+    ----------
+    f0 : ndarray
+        Input F0 contour.
+    spectrogram : ndarray
+        Spectral envelope.
+    aperiodicity : ndarray
+        Aperodicity envelope.
+    fs : int
+        Sample rate of input signal in Hz.
+    frame_period : float
+        Period between consecutive frames in milliseconds.
+        Default: 5.0
+
+    Returns
+    -------
+    y : ndarray
+        Output waveform signal.
+    """
+    ...
+
+def get_num_aperiodicities(fs: int) -> int:
+    """Calculate the required dimensionality to code D4C aperiodicity.
+
+    Parameters
+    ----------
+    fs : int
+        Sample rate of input signal in Hz.
+
+    Returns
+    -------
+    n_aper : int
+        Required number of coefficients.
+    """
+    ...
+
+def code_aperiodicity(aperiodicity: np.ndarray, fs: int) -> np.ndarray:
+    """Reduce dimensionality of D4C aperiodicity.
+
+    Parameters
+    ----------
+    aperiodicity : ndarray
+        Aperodicity envelope.
+    fs : int
+        Sample rate of input signal in Hz.
+
+    Returns
+    -------
+    coded_aperiodicity : ndarray
+        Coded aperiodicity envelope.
+    """
+    ...
+
+def decode_aperiodicity(
+    coded_aperiodicity: np.ndarray, fs: int, fft_size: int
+) -> np.ndarray:
+    """Restore full dimensionality of coded D4C aperiodicity.
+
+    Parameters
+    ----------
+    coded_aperiodicity : ndarray
+        Coded aperodicity envelope.
+    fs : int
+        Sample rate of input signal in Hz.
+    fft_size : int
+        FFT size corresponding to the full dimensional aperiodicity.
+
+    Returns
+    -------
+    aperiodicity : ndarray
+        Aperiodicity envelope.
+    """
+    ...
+
+def code_spectral_envelope(
+    spectrogram: np.ndarray, fs: int, number_of_dimensions: int
+) -> np.ndarray:
+    """Reduce dimensionality of spectral envelope.
+
+    Parameters
+    ----------
+    spectrogram : ndarray
+        Spectral envelope.
+    fs : int
+        Sample rate of input signal in Hz.
+    number_of_dimensions : int
+        Number of dimentions of coded spectral envelope
+
+    Returns
+    -------
+    coded_spectral_envelope : ndarray
+        Coded spectral envelope.
+    """
+    ...
+
+def decode_spectral_envelope(
+    coded_spectral_envelope: np.ndarray, fs: int, fft_size: int
+) -> np.ndarray:
+    """Restore full dimensionality of coded spectral envelope.
+
+    Parameters
+    ----------
+    coded_spectral_envelope : ndarray
+        Coded spectral envelope.
+    fs : int
+        Sample rate of input signal in Hz.
+    fft_size : int
+        FFT size corresponding to the full dimensional spectral envelope.
+
+    Returns
+    -------
+    spectrogram : ndarray
+        Spectral envelope.
+    """
+    ...
+
+def wav2world(
+    x: np.ndarray, fs: int, fft_size: Optional[int] = ..., frame_period: float = ...
+) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
+    """Convenience function to do all WORLD analysis steps in a single call.
+
+    In this case only `frame_period` can be configured and other parameters
+    are fixed to their defaults. Likewise, F0 estimation is fixed to
+    DIO plus StoneMask refinement.
+
+    Parameters
+    ----------
+    x : ndarray
+        Input waveform signal.
+    fs : int
+        Sample rate of input signal in Hz.
+    fft_size : int
+        Length of Fast Fourier Transform (in number of samples)
+        The resulting dimension of `ap` adn `sp` will be `fft_size` // 2 + 1
+    frame_period : float
+        Period between consecutive frames in milliseconds.
+        Default: 5.0
+
+    Returns
+    -------
+    f0 : ndarray
+        F0 contour.
+    sp : ndarray
+        Spectral envelope.
+    ap : ndarray
+        Aperiodicity.
+    """
+    ...
diff --git a/pyworld/pyworld.pyx b/pyworld/pyworld.pyx
index e81089c..4cbe93b 100644
--- a/pyworld/pyworld.pyx
+++ b/pyworld/pyworld.pyx
@@ -1,4 +1,5 @@
 from __future__ import division
+cimport cython
 import cython
 
 import numpy as np
@@ -11,7 +12,7 @@ cdef extern from "world/synthesis.h":
         int f0_length, const double * const *spectrogram,
         const double * const *aperiodicity,
         int fft_size, double frame_period,
-        int fs, int y_length, double *y) except + nogil
+        int fs, int y_length, double *y) nogil except +
 
 
 cdef extern from "world/cheaptrick.h":
@@ -25,7 +26,7 @@ cdef extern from "world/cheaptrick.h":
     void InitializeCheapTrickOption(int fs, CheapTrickOption *option) except +
     void CheapTrick(const double *x, int x_length, int fs, const double *temporal_positions,
         const double *f0, int f0_length, const CheapTrickOption *option,
-        double **spectrogram) except + nogil
+        double **spectrogram) nogil except +
 
 
 cdef extern from "world/dio.h":
@@ -40,7 +41,7 @@ cdef extern from "world/dio.h":
     void InitializeDioOption(DioOption *option) except +
     int GetSamplesForDIO(int fs, int x_length, double frame_period)
     void Dio(const double *x, int x_length, int fs, const DioOption *option,
-        double *temporal_positions, double *f0) except + nogil
+        double *temporal_positions, double *f0) nogil except +
 
 
 cdef extern from "world/harvest.h":
@@ -52,7 +53,7 @@ cdef extern from "world/harvest.h":
     void InitializeHarvestOption(HarvestOption *option)
     int GetSamplesForHarvest(int fs, int x_length, double frame_period)
     void Harvest(const double *x, int x_length, int fs, const HarvestOption *option,
-        double *temporal_positions, double *f0) except + nogil
+        double *temporal_positions, double *f0) nogil except +
 
 
 cdef extern from "world/d4c.h":
@@ -62,13 +63,13 @@ cdef extern from "world/d4c.h":
     void InitializeD4COption(D4COption *option) except +
     void D4C(const double *x, int x_length, int fs, const double *temporal_positions,
         const double *f0, int f0_length, int fft_size, const D4COption *option,
-        double **aperiodicity) except + nogil
+        double **aperiodicity) nogil except +
 
 
 cdef extern from "world/stonemask.h":
     void StoneMask(const double *x, int x_length, int fs,
         const double *temporal_positions, const double *f0, int f0_length,
-        double *refined_f0) except + nogil
+        double *refined_f0) nogil except +
 
 
 cdef extern from "world/codec.h":
@@ -147,7 +148,7 @@ def dio(np.ndarray[double, ndim=1, mode="c"] x not None, int fs,
         np.zeros(f0_length, dtype=np.dtype('float64'))
     cdef np.ndarray[double, ndim=1, mode="c"] temporal_positions = \
         np.zeros(f0_length, dtype=np.dtype('float64'))
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         Dio(&x[0], x_length, fs, &option, &temporal_positions[0], &f0[0])
     return f0, temporal_positions
 
@@ -191,7 +192,7 @@ def harvest(np.ndarray[double, ndim=1, mode="c"] x not None, int fs,
         np.zeros(f0_length, dtype=np.dtype('float64'))
     cdef np.ndarray[double, ndim=1, mode="c"] temporal_positions = \
         np.zeros(f0_length, dtype=np.dtype('float64'))
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         Harvest(&x[0], x_length, fs, &option, &temporal_positions[0], &f0[0])
     return f0, temporal_positions
 
@@ -222,7 +223,7 @@ def stonemask(np.ndarray[double, ndim=1, mode="c"] x not None,
     cdef int f0_length = <int>len(f0)
     cdef np.ndarray[double, ndim=1, mode="c"] refined_f0 = \
         np.zeros(f0_length, dtype=np.dtype('float64'))
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         StoneMask(&x[0], x_length, fs, &temporal_positions[0],
             &f0[0], f0_length, &refined_f0[0])
     return refined_f0
@@ -322,7 +323,7 @@ def cheaptrick(np.ndarray[double, ndim=1, mode="c"] x not None,
     cdef np.intp_t[:] tmp = np.zeros(f0_length, dtype=np.intp)
     cdef double **cpp_spectrogram = <double**> (<void*> &tmp[0])
     cdef np.intp_t i
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         for i in range(f0_length):
             cpp_spectrogram[i] = &spectrogram[i, 0]
 
@@ -390,7 +391,7 @@ def d4c(np.ndarray[double, ndim=1, mode="c"] x not None,
     cdef np.intp_t[:] tmp = np.zeros(f0_length, dtype=np.intp)
     cdef double **cpp_aperiodicity = <double**> (<void*> &tmp[0])
     cdef np.intp_t i
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         for i in range(f0_length):
             cpp_aperiodicity[i] = &aperiodicity[i, 0]
 
@@ -450,7 +451,7 @@ def synthesize(np.ndarray[double, ndim=1, mode="c"] f0 not None,
     cdef double **cpp_spectrogram = <double**> (<void*> &tmp[0])
     cdef double **cpp_aperiodicity = <double**> (<void*> &tmp2[0])
     cdef np.intp_t i
-    with (nogil, cython.boundscheck(False)):
+    with nogil, cython.boundscheck(False):
         for i in range(f0_length):
             cpp_spectrogram[i] = &spectrogram0[i, 0]
             cpp_aperiodicity[i] = &aperiodicity0[i, 0]
diff --git a/requirements.txt b/requirements.txt
index c397591..329bcdb 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,4 +1,2 @@
-numpy
-matplotlib
-argparse; python_version<"3.5"
-soundfile
+numpy
+argparse; python_version<"3.5"
diff --git a/setup.py b/setup.py
index d60ce23..77255e2 100644
--- a/setup.py
+++ b/setup.py
@@ -9,7 +9,7 @@
 from setuptools.command.build_ext import build_ext
 
 
-_VERSION = '0.3.4'
+_VERSION = '0.3.6'
 
 
 world_src_top = join("lib", "World", "src")
@@ -32,7 +32,11 @@
     cmdclass={'build_ext': build_ext},
     version=_VERSION,
     packages=find_packages(),
-    install_requires=['numpy'],
+    package_data={"pyworld": ["py.typed", "*.pyi"]},
+    install_requires=[
+        'numpy',
+        'importlib-metadata; python_version<"3.8"',
+    ],
     extras_require={
         'test': ['nose'],
         'sdist': ['numpy', 'cython>=0.24'],