forked from mrq/DL-Art-School
Use slaney norm in the mel filterbank computation
This commit is contained in:
parent
b2d8fbcfc0
commit
faf55684b8
|
@ -607,7 +607,8 @@ class TorchMelSpectrogramInjector(Injector):
|
|||
self.mel_stft = torchaudio.transforms.MelSpectrogram(n_fft=self.filter_length, hop_length=self.hop_length,
|
||||
win_length=self.win_length, power=2, normalized=norm,
|
||||
sample_rate=self.sampling_rate, f_min=self.mel_fmin,
|
||||
f_max=self.mel_fmax, n_mels=self.n_mel_channels)
|
||||
f_max=self.mel_fmax, n_mels=self.n_mel_channels,
|
||||
norm="slaney")
|
||||
|
||||
def forward(self, state):
|
||||
inp = state[self.input]
|
||||
|
|
Loading…
Reference in New Issue
Block a user