They would be super useful as fundamental ops (and autograd ops). STFT can be implemented in terms of ConvTranspose, see github.com/pseeth/pytorch-stft - [x] stft - [x] istft - [x] window functions: bartlett, hann, hamming, blackman - [x] fft, ifft, irfft, rfft - [x] fftshift, ifftshift cc @mruberry
