def __getitem__(self, idx):
ys, meta, _, xs = self.mix[idx]
xs = torch.from_numpy(xs)
ys = torch.from_numpy(ys)
X1s = torchaudio.functional.spectrogram(waveform=torch.transpose(
xs[0, :, :], 0, 1),
pad=0,
window=torch.hann_window(400),
n_fft=512,
hop_length=256,
win_length=400,
power=None,
normalized=False)
X2s = torchaudio.functional.spectrogram(waveform=torch.transpose(
xs[1, :, :], 0, 1),
pad=0,
window=torch.hann_window(400),
n_fft=512,
hop_length=256,
win_length=400,
power=None,
normalized=False)
Ys = torchaudio.functional.spectrogram(waveform=torch.transpose(
ys, 0, 1),
pad=0,
window=torch.hann_window(400),
n_fft=512,
hop_length=256,
win_length=400,
power=None,
normalized=False)
M1s = (X1s[0, :, :, 0]**2 +
X1s[0, :, :, 1]**2) / (X1s[0, :, :, 0]**2 + X1s[0, :, :, 1]**2 +
X2s[0, :, :, 0]**2 + X2s[0, :, :, 1]**2)
M2s = (X1s[1, :, :, 0]**2 +
X1s[1, :, :, 1]**2) / (X1s[1, :, :, 0]**2 + X1s[1, :, :, 1]**2 +
X2s[1, :, :, 0]**2 + X2s[1, :, :, 1]**2)
Ms = torch.cat(
(torch.unsqueeze(M1s, dim=0), torch.unsqueeze(M2s, dim=0)), 0)
tau = BIRD.getTDOA(meta)[0]
return Ys, Ms, tau
def __getitem__(self, idx):
ys, meta, _, _ = self.mix[idx]
ys = torch.from_numpy(ys)
Ys = torchaudio.functional.spectrogram(waveform=torch.transpose(
ys, 0, 1),
pad=0,
window=torch.hann_window(400),
n_fft=512,
hop_length=256,
win_length=400,
power=None,
normalized=False)
taus = BIRD.getTDOA(meta)
taus = torch.tensor(taus[0:2])
return Ys, taus