testing with sox only when sox is available (#419)

* testing with sox only when sox is available.

* use wav instead of mp3 for testing functions.

* typo.

* guard against not sox.

* backends definition.

* grouping backend functions into a separate file.

* remove duplicated code.

* requires sox.

* replace by wav, requires sox.

* require with scope.

* undo alignment.

* requires sox for these two, because of mp3.

* no longer need first mp3.

* cleaning.

* new wav version of file.

* flake8.

* remove unnecessary load.

* flake8.

* lint.

* lint.

* revert formatting of file.

* merging into common_utils.

* docstring.

* rename to common_utils.

test/common_utils.py

@@ -1,10 +1,13 @@
 import os
-from shutil import copytree
 import tempfile
+from contextlib import contextmanager
+from shutil import copytree
 
 import torch
+import torchaudio
 
 TEST_DIR_PATH = os.path.dirname(os.path.realpath(__file__))
+BACKENDS = torchaudio._backend._audio_backends
 
 
 def create_temp_assets_dir():
@@ -48,3 +51,35 @@ def random_int_tensor(seed, size, low=0, high=2 ** 32, a=22695477, c=1, m=2 ** 3
     """ Same as random_float_tensor but integers between [low, high)
     """
     return torch.floor(random_float_tensor(seed, size, a, c, m) * (high - low)) + low
+
+
+@contextmanager
+def AudioBackendScope(new_backend):
+    previous_backend = torchaudio.get_audio_backend()
+    try:
+        torchaudio.set_audio_backend(new_backend)
+        yield
+    finally:
+        torchaudio.set_audio_backend(previous_backend)
+
+
+def filter_backends_with_mp3(backends):
+    # Filter out backends that do not support mp3
+
+    test_dirpath, _ = create_temp_assets_dir()
+    test_filepath = os.path.join(
+        test_dirpath, "assets", "steam-train-whistle-daniel_simon.mp3"
+    )
+
+    def supports_mp3(backend):
+        try:
+            with AudioBackendScope(backend):
+                torchaudio.load(test_filepath)
+            return True
+        except RuntimeError:
+            return False
+
+    return [backend for backend in backends if supports_mp3(backend)]
+
+
+BACKENDS_MP3 = filter_backends_with_mp3(BACKENDS)

test/test.py

@@ -1,39 +1,25 @@
 import unittest
-import common_utils
 import torch
 import torchaudio
 import math
 import os
-
-
-class AudioBackendScope:
-    def __init__(self, backend):
-        self.new_backend = backend
-        self.previous_backend = torchaudio.get_audio_backend()
-
-    def __enter__(self):
-        torchaudio.set_audio_backend(self.new_backend)
-        return self.new_backend
-
-    def __exit__(self, type, value, traceback):
-        backend = self.previous_backend
-        torchaudio.set_audio_backend(backend)
+from common_utils import AudioBackendScope, BACKENDS, BACKENDS_MP3, create_temp_assets_dir
 
 
 class Test_LoadSave(unittest.TestCase):
-    test_dirpath, test_dir = common_utils.create_temp_assets_dir()
+    test_dirpath, test_dir = create_temp_assets_dir()
     test_filepath = os.path.join(test_dirpath, "assets",
                                  "steam-train-whistle-daniel_simon.mp3")
     test_filepath_wav = os.path.join(test_dirpath, "assets",
                                      "steam-train-whistle-daniel_simon.wav")
 
     def test_1_save(self):
-        for backend in ["sox"]:
+        for backend in BACKENDS_MP3:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_1_save(self.test_filepath, False)
 
-        for backend in ["sox", "soundfile"]:
+        for backend in BACKENDS:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_1_save(self.test_filepath_wav, True)
@@ -79,7 +65,7 @@ def _test_1_save(self, test_filepath, normalization):
             torchaudio.save(new_filepath, x, sr)
 
     def test_1_save_sine(self):
-        for backend in ["sox", "soundfile"]:
+        for backend in BACKENDS:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_1_save_sine()
@@ -112,12 +98,12 @@ def _test_1_save_sine(self):
         os.unlink(new_filepath)
 
     def test_2_load(self):
-        for backend in ["sox"]:
+        for backend in BACKENDS_MP3:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_2_load(self.test_filepath, 278756)
 
-        for backend in ["sox", "soundfile"]:
+        for backend in BACKENDS:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_2_load(self.test_filepath_wav, 276858)
@@ -153,7 +139,7 @@ def _test_2_load(self, test_filepath, length):
             torchaudio.load(tdir)
 
     def test_2_load_nonormalization(self):
-        for backend in ["sox"]:
+        for backend in BACKENDS_MP3:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_2_load_nonormalization(self.test_filepath, 278756)
@@ -170,7 +156,7 @@ def _test_2_load_nonormalization(self, test_filepath, length):
         self.assertTrue(isinstance(x, torch.LongTensor))
 
     def test_3_load_and_save_is_identity(self):
-        for backend in ["sox", "soundfile"]:
+        for backend in BACKENDS:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_3_load_and_save_is_identity()
@@ -185,6 +171,7 @@ def _test_3_load_and_save_is_identity(self):
         self.assertEqual(sample_rate, sample_rate2)
         os.unlink(output_path)
 
+    @unittest.skipIf(set(["sox", "soundfile"]) not in set(BACKENDS), "sox and soundfile are not available")
     def test_3_load_and_save_is_identity_across_backend(self):
         with self.subTest():
             self._test_3_load_and_save_is_identity_across_backend("sox", "soundfile")
@@ -208,7 +195,7 @@ def _test_3_load_and_save_is_identity_across_backend(self, backend1, backend2):
         os.unlink(output_path)
 
     def test_4_load_partial(self):
-        for backend in ["sox"]:
+        for backend in BACKENDS_MP3:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_4_load_partial()
@@ -250,7 +237,7 @@ def _test_4_load_partial(self):
             torchaudio.load(input_sine_path, offset=100000)
 
     def test_5_get_info(self):
-        for backend in ["sox", "soundfile"]:
+        for backend in BACKENDS:
             with self.subTest():
                 with AudioBackendScope(backend):
                     self._test_5_get_info()

test/test_compliance_kaldi.py

@@ -1,11 +1,11 @@
 import math
 import os
-import common_utils
 import compliance.utils
 import torch
 import torchaudio
 import torchaudio.compliance.kaldi as kaldi
 import unittest
+from common_utils import AudioBackendScope, BACKENDS, create_temp_assets_dir
 
 
 def extract_window(window, wave, f, frame_length, frame_shift, snip_edges):
@@ -45,7 +45,7 @@ def first_sample_of_frame(frame, window_size, window_shift, snip_edges):
 
 
 class Test_Kaldi(unittest.TestCase):
-    test_dirpath, test_dir = common_utils.create_temp_assets_dir()
+    test_dirpath, test_dir = create_temp_assets_dir()
     test_filepath = os.path.join(test_dirpath, 'assets', 'kaldi_file.wav')
     test_8000_filepath = os.path.join(test_dirpath, 'assets', 'kaldi_file_8000.wav')
     kaldi_output_dir = os.path.join(test_dirpath, 'assets', 'kaldi')
@@ -159,6 +159,8 @@ def _compliance_test_helper(self, sound_filepath, filepath_key, expected_num_fil
             self.assertTrue(output.shape, kaldi_output.shape)
             self.assertTrue(torch.allclose(output, kaldi_output, atol=atol, rtol=rtol))
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_spectrogram(self):
         def get_output_fn(sound, args):
             output = kaldi.spectrogram(
@@ -179,6 +181,8 @@ def get_output_fn(sound, args):
 
         self._compliance_test_helper(self.test_filepath, 'spec', 131, 13, get_output_fn, atol=1e-3, rtol=0)
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_fbank(self):
         def get_output_fn(sound, args):
             output = kaldi.fbank(
@@ -209,6 +213,8 @@ def get_output_fn(sound, args):
 
         self._compliance_test_helper(self.test_filepath, 'fbank', 97, 22, get_output_fn, atol=1e-3, rtol=1e-1)
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_mfcc(self):
         def get_output_fn(sound, args):
             output = kaldi.mfcc(
@@ -243,6 +249,8 @@ def test_mfcc_empty(self):
         # Passing in an empty tensor should result in an error
         self.assertRaises(AssertionError, kaldi.mfcc, torch.empty(0))
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_resample_waveform(self):
         def get_output_fn(sound, args):
             output = kaldi.resample_waveform(sound, args[1], args[2])

test/test_dataloader.py

@@ -1,16 +1,17 @@
 import unittest
-import common_utils
 import torch
 import torch.nn as nn
 from torch.utils.data import Dataset, DataLoader
 import torchaudio
 import math
 import os
+from common_utils import AudioBackendScope, BACKENDS, create_temp_assets_dir
 
 
+@unittest.skipIf("sox" not in BACKENDS, "sox not available")
 class TORCHAUDIODS(Dataset):
 
-    test_dirpath, test_dir = common_utils.create_temp_assets_dir()
+    test_dirpath, test_dir = create_temp_assets_dir()
 
     def __init__(self):
         self.asset_dirpath = os.path.join(self.test_dirpath, "assets")
@@ -33,6 +34,7 @@ def __len__(self):
         return len(self.data)
 
 
+@unittest.skipIf("sox" not in BACKENDS, "sox not available")
 class Test_DataLoader(unittest.TestCase):
     @classmethod
     def setUpClass(cls):
@@ -50,4 +52,5 @@ def test_1(self):
             self.assertTrue(x.size() == expected_size)
 
 if __name__ == '__main__':
-    unittest.main()
+    with AudioBackendScope("sox"):
+        unittest.main()

test/test_functional.py

@@ -9,6 +9,7 @@
 import unittest
 import common_utils
 
+from common_utils import AudioBackendScope, BACKENDS
 from torchaudio.common_utils import IMPORT_LIBROSA
 
 if IMPORT_LIBROSA:
@@ -39,7 +40,7 @@ class TestFunctional(unittest.TestCase):
     test_dirpath, test_dir = common_utils.create_temp_assets_dir()
 
     test_filepath = os.path.join(test_dirpath, 'assets',
-                                 'steam-train-whistle-daniel_simon.mp3')
+                                 'steam-train-whistle-daniel_simon.wav')
     waveform_train, sr_train = torchaudio.load(test_filepath)
 
     def test_torchscript_spectrogram(self):
@@ -433,6 +434,8 @@ def test_create_fb(self):
         self._test_create_fb(n_mels=56, fmin=1900.0, fmax=900.0)
         self._test_create_fb(n_mels=10, fmin=1900.0, fmax=900.0)
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_gain(self):
         waveform_gain = F.gain(self.waveform_train, 3)
         self.assertTrue(waveform_gain.abs().max().item(), 1.)
@@ -444,6 +447,8 @@ def test_gain(self):
 
         self.assertTrue(torch.allclose(waveform_gain, sox_gain_waveform, atol=1e-04))
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_dither(self):
         waveform_dithered = F.dither(self.waveform_train)
         waveform_dithered_noiseshaped = F.dither(self.waveform_train, noise_shaping=True)
@@ -461,6 +466,8 @@ def test_dither(self):
 
         self.assertTrue(torch.allclose(waveform_dithered_noiseshaped, sox_dither_waveform_ns, atol=1e-02))
 
+    @unittest.skipIf("sox" not in BACKENDS, "sox not available")
+    @AudioBackendScope("sox")
     def test_vctk_transform_pipeline(self):
         test_filepath_vctk = os.path.join(self.test_dirpath, "assets/VCTK-Corpus/wav48/p224/", "p224_002.wav")
         wf_vctk, sr_vctk = torchaudio.load(test_filepath_vctk)