In speech processing , why we do windowing , and why we must choose the window to be smooth(not triangular)?