def test_forward():
layer = LogMel(n_fft=16, n_mels=2)
x = torch.randn(2, 4, 9)
y, _ = layer(x)
assert y.shape == (2, 4, 2)
y, ylen = layer(x, torch.tensor([4, 2], dtype=torch.long))
assert (ylen == torch.tensor((4, 2), dtype=torch.long)).all()
def __init__(
self,
fs: Union[int, str] = 16000,
n_fft: int = 512,
win_length: int = None,
hop_length: int = 128,
window: Optional[str] = "hann",
center: bool = True,
normalized: bool = False,
onesided: bool = True,
n_mels: int = 80,
fmin: int = None,
fmax: int = None,
htk: bool = False,
frontend_conf: Optional[dict] = get_default_kwargs(Frontend),
apply_stft: bool = True,
):
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
# Deepcopy (In general, dict shouldn't be used as default arg)
frontend_conf = copy.deepcopy(frontend_conf)
self.hop_length = hop_length
if apply_stft:
self.stft = Stft(
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
center=center,
window=window,
normalized=normalized,
onesided=onesided,
)
else:
self.stft = None
self.apply_stft = apply_stft
if frontend_conf is not None:
self.frontend = Frontend(idim=n_fft // 2 + 1, **frontend_conf)
else:
self.frontend = None
self.logmel = LogMel(
fs=fs,
n_fft=n_fft,
n_mels=n_mels,
fmin=fmin,
fmax=fmax,
htk=htk,
)
self.n_mels = n_mels
self.frontend_type = "default"
def __init__(
self,
fs: Union[int, str] = 16000,
n_fft: int = 1024,
win_length: int = None,
hop_length: int = 256,
window: Optional[str] = "hann",
center: bool = True,
pad_mode: str = "reflect",
normalized: bool = False,
onesided: bool = True,
n_mels: int = 80,
fmin: Optional[int] = 80,
fmax: Optional[int] = 7600,
htk: bool = False,
):
assert check_argument_types()
super().__init__()
if isinstance(fs, str):
fs = humanfriendly.parse_size(fs)
self.fs = fs
self.n_mels = n_mels
self.n_fft = n_fft
self.hop_length = hop_length
self.win_length = win_length
self.window = window
self.fmin = fmin
self.fmax = fmax
self.stft = Stft(
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
window=window,
center=center,
pad_mode=pad_mode,
normalized=normalized,
onesided=onesided,
)
self.logmel = LogMel(
fs=fs,
n_fft=n_fft,
n_mels=n_mels,
fmin=fmin,
fmax=fmax,
htk=htk,
log_base=10.0,
)
def test_repr():
print(LogMel())
def test_backward_not_leaf_in():
layer = LogMel(n_fft=16, n_mels=2)
x = torch.randn(2, 4, 9, requires_grad=True)
x = x + 2
y, _ = layer(x)
y.sum().backward()