Merge pull request #133 from r9y9/np2

relax numpy requirement and support numpy 2.0.0

.github/workflows/ci.yaml

@@ -50,11 +50,20 @@ jobs:
             brew install libsamplerate
             ;;
           esac
-    - name: Install dependencies
+    - name: Install dependencies (Linux and macOS)
+      if: runner.os != 'Windows'
       run: |
         python -m pip install --upgrade pip
         pip install torch
         pip install -e ".[test,lint]"
+    # NOTE: window is not happy with numpy 2.0.0
+    - name: Install dependencies (Windows)
+      if: runner.os == 'Windows'
+      run: |
+        python -m pip install --upgrade pip
+        pip install "numpy<2.0.0"
+        pip install torch
+        pip install -e ".[test,lint]"
     - name: Lint with pysen
       run: |
         pysen run lint

pyproject.toml

@@ -3,7 +3,7 @@ requires = [
     "wheel",
     "setuptools",
     "cython>=0.21.0",
-    "numpy>=v1.20.0, <2",
+    "numpy>=v1.20.0",
     "scipy",
 ]
 

setup.py

@@ -140,7 +140,7 @@ def package_files(directory):
     "scikit-learn",
     "pysptk >= 0.1.17",
     "tqdm",
-    "numpy >= 1.20.0, <2",
+    "numpy >= 1.20.0",
 ]
 
 setup(

tests/test_preprocessing.py

@@ -1,8 +1,6 @@
-import librosa
 import numpy as np
 import pysptk
 import pytest
-import pyworld
 from nnmnkwii import preprocessing as P
 from nnmnkwii.datasets import FileSourceDataset, PaddedFileSourceDataset
 from nnmnkwii.preprocessing import (
@@ -20,8 +18,17 @@
     example_file_data_sources_for_acoustic_model,
     example_file_data_sources_for_duration_model,
 )
+from packaging.version import Version
 from scipy.io import wavfile
 
+try:
+    import pyworld
+
+    pyworld_available = True
+except ValueError:
+    # ValueError: numpy.dtype size changed, may indicate binary incompatibility.
+    pyworld_available = False
+
 
 def _get_windows_set_bandmat():
     windows_set = [
@@ -271,6 +278,7 @@ def test_interp1d():
     assert np.all(if0 != 0)
 
 
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_trim_remove_zeros_frames():
     fs, x = wavfile.read(example_audio_file())
     frame_period = 5
@@ -449,12 +457,19 @@ def test_dtw_frame_length_adjustment():
         assert X_aligned.shape == Y_aligned.shape
 
 
+@pytest.mark.skipif(
+    Version(np.__version__) >= Version("2.0.0"), reason="numpy >= 2.0.0"
+)
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_dtw_aligner():
     from nnmnkwii.preprocessing.alignment import DTWAligner, IterativeDTWAligner
 
     fs, x = wavfile.read(example_audio_file())
     x = (x / 32768.0).astype(np.float32)
     assert fs == 16000
+    # NOTE: librosa deps need to be updated for numpy 2.0.0
+    import librosa
+
     x_fast = librosa.effects.time_stretch(x, rate=2.0)
 
     X = _get_mcep(x, fs)

tests/test_real_datasets.py

@@ -5,7 +5,6 @@
 import numpy as np
 import pysptk
 import pytest
-import pyworld
 
 # Data source implementations
 from nnmnkwii.datasets import (
@@ -21,7 +20,13 @@
 from nnmnkwii.preprocessing import trim_zeros_frames
 from scipy.io import wavfile
 
-# Tests marked with "require_local_data" needs data to be downloaded.
+try:
+    import pyworld
+
+    pyworld_available = True
+except ValueError:
+    # ValueError: numpy.dtype size changed, may indicate binary incompatibility.
+    pyworld_available = False
 
 
 def test_cmu_arctic_dummy():
@@ -142,6 +147,7 @@ def f(source):
 @pytest.mark.skipif(
     not (Path.home() / "data" / "cmu_arctic").exists(), reason="Data doesn't exist"
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_cmu_arctic():
     DATA_DIR = join(expanduser("~"), "data", "cmu_arctic")
     if not exists(DATA_DIR):
@@ -205,6 +211,7 @@ def collect_features(self, path):
     not (Path.home() / "data" / "voice-statistics").exists(),
     reason="Data doesn't exist",
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_voice_statistics():
     DATA_DIR = join(expanduser("~"), "data", "voice-statistics")
     if not exists(DATA_DIR):
@@ -303,6 +310,7 @@ def collect_features(self, path):
 @pytest.mark.skipif(
     not (Path.home() / "data" / "LJSpeech-1.1").exists(), reason="Data doesn't exist"
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_ljspeech():
     DATA_DIR = join(expanduser("~"), "data", "LJSpeech-1.1")
     if not exists(DATA_DIR):
@@ -355,6 +363,7 @@ def collect_features(self, path):
 @pytest.mark.skipif(
     not (Path.home() / "data" / "vcc2016").exists(), reason="Data doesn't exist"
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_vcc2016():
     DATA_DIR = join(expanduser("~"), "data", "vcc2016")
     if not exists(DATA_DIR):
@@ -417,6 +426,7 @@ def collect_features(self, path):
 @pytest.mark.skipif(
     not (Path.home() / "data" / "jsut_ver1.1").exists(), reason="Data doesn't exist"
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_jsut():
     DATA_DIR = join(expanduser("~"), "data", "jsut_ver1.1")
     if not exists(DATA_DIR):
@@ -475,6 +485,7 @@ def collect_features(self, path):
 @pytest.mark.skipif(
     not (Path.home() / "data" / "VCTK-Corpus").exists(), reason="Data doesn't exist"
 )
+@pytest.mark.skipif(not pyworld_available, reason="pyworld is not available")
 def test_vctk():
     DATA_DIR = join(expanduser("~"), "data", "VCTK-Corpus")
     if not exists(DATA_DIR):