def assert_sphere(
self,
dtype,
sample_rate,
num_channels,
channels_first=True,
duration=1,
):
"""`soundfile_backend.load` can load SPHERE format correctly."""
path = self.get_temp_path("reference.sph")
num_frames = duration * sample_rate
raw = get_wav_data(
dtype,
num_channels,
num_frames=num_frames,
normalize=False,
channels_first=False,
)
soundfile.write(path,
raw,
sample_rate,
subtype=dtype2subtype(dtype),
format="NIST")
expected = normalize_wav(raw.t() if channels_first else raw)
data, sr = soundfile_backend.load(path, channels_first=channels_first)
assert sr == sample_rate
self.assertEqual(data, expected, atol=1e-4, rtol=1e-8)
def assert_wav(
self,
dtype,
sample_rate,
num_channels,
normalize,
channels_first=True,
duration=1,
):
"""`soundfile_backend.load` can load wav format correctly.
Wav data loaded with soundfile backend should match those with scipy
"""
path = self.get_temp_path("reference.wav")
num_frames = duration * sample_rate
data = get_wav_data(
dtype,
num_channels,
normalize=normalize,
num_frames=num_frames,
channels_first=channels_first,
)
save_wav(path, data, sample_rate, channels_first=channels_first)
expected = load_wav(path,
normalize=normalize,
channels_first=channels_first)[0]
data, sr = soundfile_backend.load(path,
normalize=normalize,
channels_first=channels_first)
assert sr == sample_rate
self.assertEqual(data, expected)
def _test_fileobj(self, ext):
"""Loading audio via file-like object works"""
sample_rate = 16000
path = self.get_temp_path(f'test.{ext}')
data = get_wav_data('float32', num_channels=2).numpy().T
soundfile.write(path, data, sample_rate)
expected = soundfile.read(path, dtype='float32')[0].T
with open(path, 'rb') as fileobj:
found, sr = soundfile_backend.load(fileobj)
assert sr == sample_rate
self.assertEqual(expected, found)
def assert_dtype(self, ext, dtype, sample_rate, num_channels, normalize,
channels_first):
"""When format is WAV or NIST, normalize=False will return the native dtype Tensor, otherwise float32"""
num_frames = 3 * sample_rate
path = _get_mock_path(ext, dtype, sample_rate, num_channels,
num_frames)
expected_dtype = (torch.float32 if normalize
or ext not in ["wav", "nist"] else getattr(
torch, dtype))
with patch("soundfile.SoundFile", SoundFileMock):
found, sr = soundfile_backend.load(path,
normalize=normalize,
channels_first=channels_first)
assert found.dtype == expected_dtype
assert sample_rate == sr
def _test_tarfile(self, ext):
"""Loading audio via file-like object works"""
sample_rate = 16000
audio_file = f'test.{ext}'
audio_path = self.get_temp_path(audio_file)
archive_path = self.get_temp_path('archive.tar.gz')
data = get_wav_data('float32', num_channels=2).numpy().T
soundfile.write(audio_path, data, sample_rate)
expected = soundfile.read(audio_path, dtype='float32')[0].T
with tarfile.TarFile(archive_path, 'w') as tarobj:
tarobj.add(audio_path, arcname=audio_file)
with tarfile.TarFile(archive_path, 'r') as tarobj:
fileobj = tarobj.extractfile(audio_file)
found, sr = soundfile_backend.load(fileobj)
assert sr == sample_rate
self.assertEqual(expected, found)
def _test_format(self, format_):
"""Providing format allows to read file without extension"""
path, expected = self._make_file(format_)
found, _ = soundfile_backend.load(path)
self.assertEqual(found, expected)
