The Spectrumworx manual is actually a great resource for this if you're yet to dig into it (no judgement implied - I know I haven't read the manuals of plenty of the stuff I use). Part 3 goes into some fairly solid detail about the fundamentals, e.g. time/frequency trade-off, plus a description of each module in later sections. Apologies if you've already read this and were looking for more.
Having a look around I found this on the CDP website: http://www.composersdesktop.com/pdf/phasevocoder.pdf- there are a few good insights (plus much IMO unnecessary detail) in there. It does a fairly good job of describing time/frequency tradeoff, why windowing is necessary, and why spectral transforms are more detailed in the upper frequencies rather than lower (plus why the lowest frequency we can capture is determined by the window size), among other things.
I'm not entirely sure what the pvd modules in Spectrumworx do myself to be honest. As far as I can tell, it switches from "amplitude/phase with equally spaced frequency bins" in the standard mode to "true frequency/amplitude with no phase information" in the pvd mode. The thing is, as far as I know, pretty much every transformation that moves the frequency bins up and down (rather than reshapes the overall contour like in analog-style filtering) means the phase information needs to be junked anyway as it no longer makes sense with the new frequencies (the same applies to time-stretching, the phases no longer make sense). SPEAR is a standalone program that lets you resynthesise with or without phase info regardless of what you've done, so playing with that is maybe a good illustration of why phase needs to be ignored in these cases. I wonder if this is why a lot of modules tend to produce a constant comb-filterish ringing when used in pvd - because the module had set the phase info to zero to begin with, so it can't find the 'true' frequencies by unwrapping the phase (as the phase vocoder does) and you just get comb-filterish sludge, the pitch of said comb-filtering determined only by the window size/overlap factor. So yeah, Spectrumworx implementation of pvd seems a little perplexing as some of the standard modules (e.g. pitch-shift) only really make sense in pvd so it seems odd to have both a pvd and non-pvd version. Sure, it means we can recover the signal exactly with no time-smearing if we've got a pitch-shift value of 0, but nobody loads a pitch-shifter to not shift pitch!
Calling the DSP gurus once more re my pvd musings!
spectral blur/spectral averaging vst for windows
-
- KVRAF
- 3477 posts since 27 Dec, 2002 from North East England
-
- KVRian
- 1146 posts since 24 Jul, 2008 from England
Many thanks for the lengthy reply, much appreciated.
I did read the manual at some point, maybe 2008, but I remember a lot of stuff going above my head. I'll have another gander and cross reference this time
Some other things you mention make sense at this point such as the frequency domain windowing. The pvd implementation does leave me scratching my head, but perhaps that's just the nature of the beast? I'm definitely no tech DSP wizard.
The sort of spectral I can wrap my head around are plugins like MSpectralDynamics, where it's presented as a more traditional tool, and demystified to where I don't need to understand under the hood too much. Deeper spectral tools feel more like a dark art, but I really appreciate what they can do.
Again, thanks for posting the info, links and trying to unveil the 'mystery'. I'll get my thinking cap on.
I did read the manual at some point, maybe 2008, but I remember a lot of stuff going above my head. I'll have another gander and cross reference this time
Some other things you mention make sense at this point such as the frequency domain windowing. The pvd implementation does leave me scratching my head, but perhaps that's just the nature of the beast? I'm definitely no tech DSP wizard.
The sort of spectral I can wrap my head around are plugins like MSpectralDynamics, where it's presented as a more traditional tool, and demystified to where I don't need to understand under the hood too much. Deeper spectral tools feel more like a dark art, but I really appreciate what they can do.
Again, thanks for posting the info, links and trying to unveil the 'mystery'. I'll get my thinking cap on.
