Preanalyse FFT ; how do I deal with the Window?

Question

I was trying to pre-analyse FFT data to disk and play it back, to explore some time stretching approaches. I capture the two channels of FFT data post analysis to disk, then play them back with variable rate into an inverse FFT.

Problem is, the Window! It needs to happen in the time-domain just before  the 2x overlap FFT and then windowed again at the iFFT stage. The iFFT is realtime, so how do I work with the window signal, which amplitude modulated the FFT data.

I tried already adjusting the rate of the iFFT window in realtime, by the same ratio as the disk based FFT data but this does not seem to work.

Any suggestions?

Comments

In the FFT prototypes the iFFT without a window has an FFT with one. And the iFFT with a window has an FFT without a window.  Why is that?

---

why would you want to change the windowing rate? the iFFT and the windwoing is determined by the FFTSize, not the playback speed. I‘m not sure if I understand the problem 100% though..

---

@Alan there is only one stage of windowing because otherwise you would apply the window twice and you wouldn‘t get a nice overlap without amplitude modulation. However if you do something to the data in-between FFT and iFFT you need to window again after the iFFT because you probably „destroyed“ the windowing of the input.

---

@kymaguy I don't want to change the windowing-rate. What I was trying to do, was analyse the FFT to disk, so that I can resynthesise a lower rate of the analysis data. The idea is to time stretch the FFT data. But I realise that if I render the analysis with a Window, the rendered data playing back slower will also alter the window position won't it?

---

@cristian-vogel you can‘t just play back the FFT data slower, you need to preserve each frame as a whole and for a stretch you need to interpolate between frames. Also you need to keep the phase running for the interpolated „in-between“ frames. Btw the fft data is not amp modulated but convolved with the window since amp modulation in the time domain results in convolution in the frequency domain and vice versa.