While extracting features of a wave tensor, I tried to use delta coefficients along with 40 filter-bank coefficients by setting deltas=True in speechbrain.lobes.features.Fbank. However, this runs into

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument weight in method wrapper___conv_depthwise2d)

The error lies in the this line.
Changing it to (just for debug shake)

torch.nn.functional.conv1d(x.to(device='cpu'), self.kernel, groups=x.shape[1])

fixes the error.
i.e. the input and kernel tensor must be put into the same device.

Thanks