Skip to content

[BC-Breaking] Remove lazy behavior from MelScale #1571

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 3 commits into from
Closed

[BC-Breaking] Remove lazy behavior from MelScale #1571

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
dbfbdce
Remove lazy behavior from MelScale
Patil2099 Jun 11, 2021
40b5a3f
Implement Feedback
Patil2099 Jun 21, 2021
9a1a116
Update torchaudio/transforms.py
mthrok Jul 14, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 0 additions & 4 deletions test/torchaudio_unittest/transforms/torchscript_consistency_impl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,10 +59,6 @@ def test_AmplitudeToDB(self):
spec = torch.rand((6, 201))
self._assert_consistency(T.AmplitudeToDB(), spec)

def test_MelScale_invalid(self):
with self.assertRaises(ValueError):
torch.jit.script(T.MelScale())

def test_MelScale(self):
spec_f = torch.rand((1, 201, 6))
self._assert_consistency(T.MelScale(n_stft=201), spec_f)
Expand Down
34 changes: 2 additions & 32 deletions torchaudio/transforms.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -255,7 +255,7 @@ def __init__(self,
sample_rate: int = 16000,
f_min: float = 0.,
f_max: Optional[float] = None,
n_stft: Optional[int] = None,
n_stft: int = 201,
norm: Optional[str] = None,
mel_scale: str = "htk") -> None:
super(MelScale, self).__init__()
Expand All @@ -267,34 +267,11 @@ def __init__(self,
self.mel_scale = mel_scale

assert f_min <= self.f_max, 'Require f_min: {} < f_max: {}'.format(f_min, self.f_max)

if n_stft is None or n_stft == 0:
warnings.warn(
'Initialization of torchaudio.transforms.MelScale with an unset weight '
'`n_stft=None` is deprecated and will be removed in release 0.10. '
'Please set a proper `n_stft` value. Typically this is `n_fft // 2 + 1`. '
'Refer to https://github.com/pytorch/audio/issues/1510 '
'for more details.'
)

fb = torch.empty(0) if n_stft is None else F.create_fb_matrix(
fb = F.create_fb_matrix(
n_stft, self.f_min, self.f_max, self.n_mels, self.sample_rate, self.norm,
self.mel_scale)
self.register_buffer('fb', fb)

def __prepare_scriptable__(self):
r"""If `self.fb` is empty, the `forward` method will try to resize the parameter,
which does not work once the transform is scripted. However, this error does not happen
until the transform is executed. This is inconvenient especially if the resulting
TorchScript object is executed in other environments. Therefore, we check the
validity of `self.fb` here and fail if the resulting TS does not work.

Returns:
MelScale: self
"""
if self.fb.numel() == 0:
raise ValueError("n_stft must be provided at construction")
return self

def forward(self, specgram: Tensor) -> Tensor:
r"""
Expand All @@ -309,13 +286,6 @@ def forward(self, specgram: Tensor) -> Tensor:
shape = specgram.size()
specgram = specgram.reshape(-1, shape[-2], shape[-1])

if self.fb.numel() == 0:
tmp_fb = F.create_fb_matrix(specgram.size(1), self.f_min, self.f_max,
self.n_mels, self.sample_rate, self.norm,
self.mel_scale)
# Attributes cannot be reassigned outside __init__ so workaround
self.fb.resize_(tmp_fb.size())
self.fb.copy_(tmp_fb)

# (channel, frequency, time).transpose(...) dot (frequency, n_mels)
# -> (channel, time, n_mels).transpose(...)
Expand Down