vichug

In the last subpatcher and paragraph : time stretching, spectral informations are recorded in a buffer~in order to re reead them at a different rate. Faire enough. But, instead of recording the phase information directly, there is a computing of phase difference between a spectral bin at a given spectral frame and the same spectral bin at next frame, using framedelta~. That value is wrapped with phasewrap~ then at last recorded.

From what i understand : a spectral frame has all the spectral bins of the spectrum, so the framedelta computes the difference between a sample at a certain position in a signal vector, and the next one at the same position in the next signal vector. Is that it ? It would be comparable to delta~, except computing between evenly positioned samples in successive signal vectors rather than successive samples, regardless of the signal vector.

Then for the sound to be played back, the phase is reconstructed, usign frameaccum~. Is that needed because sometimes one fft frame is played more than once, or a fft frame can be skipped, depending on playback speed, and the phase needs to be evolving continuously from the previous one ?

I hope i explained ok what i think i have understood, any further explanation will be welcome !

Hey,
In the last subpatcher and paragraph : time stretching, spectral informations are recorded in a buffer~in order to re reead them at a different rate. Faire enough. But, instead of recording the phase information directly, there is a computing of phase difference between a spectral bin at a given spectral frame and the same spectral bin at next frame, using framedelta~. That value is wrapped with phasewrap~ then at last recorded.
From what i understand : a spectral frame has all the spectral bins of the spectrum, so the framedelta computes the difference between a sample at a certain position in a signal vector, and the next one at the same position in the next signal vector. Is that it ? It would be comparable to delta~, except computing between evenly positioned samples in successive signal vectors rather than successive samples, regardless of the signal vector.
Then for the sound to be played back, the phase is reconstructed, usign frameaccum~. Is that needed because sometimes one fft frame is played more than once, or a fft frame can be skipped, depending on playback speed, and the phase needs to be evolving continuously from the previous one ?
I hope i explained ok what i think i have understood, any further explanation will be welcome !


understanding-msp-tutorial-26-pfft-frequency-domain-signal-processing

If I understand correctly what you think you understood, I think you understood more or less correctly. :)

Here's the dude, Mark Dolson, on the topic. See especially page 23.

If I understand correctly what you think you understood, I think you understood more or less correctly. :)
Here's the dude, Mark Dolson, on the topic. See especially page 23.The Phase Vocoder: A Tutorial


understanding msp tutorial 26 : PFFT~, frequency domain signal processing