Resample supports only one channel

[vincentqb]

torchaudio.transforms.Resample appears to only apply to single channels.

tensor.size()
# torch.Size([2, 276858])

torchaudio.transforms.Resample(frequency, new_frequency)(tensor).size()
# RuntimeError: Given groups=1, weight of size 1 1 121, expected input[1, 2, 276972] to have 1 channels, but got 2 channels instead

torchaudio.transforms.Resample(frequency, new_frequency)(tensor[0,:].view(1,-1))
# torch.Size([1, 27686])

torchaudio.transforms.Compose([
    torchaudio.transforms.LC2CL(),
    torchaudio.transforms.DownmixMono(),
    torchaudio.transforms.LC2CL(),
    torchaudio.transforms.Resample(frequency, new_frequency)
])(tensor).size()
# torch.Size([1, 27686])

Since we are in the process of standardizing #152, would it make sense to apply the resample on the last dimension, assumed to be time?

[vincentqb]

@jamarshon -- thoughts?

[jamarshon]

Fixed in #154