Phase Vocoder Circular Shift

Hi all,

I'm putting together a phase vocoder as cool little audio project and I'm starting out with time-stretching.

I have noticed in most of the examples and literature people suggest performing a circular shift of the input windows before the FFT to ensure that the phase properly taken care of. This is instead of just taking the FFT of the windowed function that has its time origin at the left-most value of the window.

Ex:

x = [1 2 3 4 5 6];

y = circshift(x, [0 length(x)/2]) = [4 5 6 1 2 3]

I am not sure why this is necessary for performing the analysis of the phase vocoder and was hoping that someone could shed some light.

Reference for use of circular shift: source.

Thanks!