My question is about the use of Hanning or Hamming windows versus "square" windows.
In my understanding, I will use a Hanning or Hamming window to limit the spectral leakage. But, on the second hand, it means I will loose the begining and the end of the signal as they are faded to zero.
I read a solution which is to "ponderate" each temporal temporal signals get by the inverted DFT as I would do with "tiles" covering each other on a roof. And, that's right, if I "recover" with a 66% ratio, my "ponderated" signal looks like the original signal multiplicated by a 1.5 ratio (excluded the 2 first and 2 last frames of my whole signal on which I can't ponderate the fading).
If I'm using that solution, it means I need to wait the 2 next samples before releasing my sample "recalculated" by the inverse DFT :
- first third of the sample n ponderated by samples n-1 and n-2 (I already got them so, I don't count them)
- second third of the sample n ponderated by samples n-1 and n+1
- last third of the sample n ponderated by samples n+1 and n+2
So, if my signal is sampled at 44.1 kHz and contains 2048 frames (to get a "good" spectral resolution), its length is approximatively 50 ms. So, my delay before releasing that signal will be 1 * 50 (=sample n) + 2 * 50 (samples n+1 and n+2) = 150 ms. It is too huge if I'm working in live conditions like in a concert.
So, my question is : why not using a square window and zero-padding BEFORE and AFTER my sample ? If I zero pad my signal before and after my original sample, won't I avoid the spectral leakage ?
Thank you for all your answers.