Peiman

What is the formula for mixing between two sounds - just a normal conventional mix, no cross synthesis - in the frequency domain, within a pfft~ subpatcher.

I'd like to do this using phase and amplitude values, rather than cartesian coordinates. This is so that I can build in a dry/wet fader within pfft~.

What is the formula for mixing between two sounds - just a normal conventional mix, no cross synthesis - in the frequency domain, within a pfft~ subpatcher. 

I'd like to do this using phase and amplitude values, rather than cartesian coordinates. This is so that I can build in a dry/wet fader within pfft~.  

fft-and-mixing-signals

Btw, Nice work on that other genwarper thing! (Also, sample-rate accumulation can be done in regular MSP, one way might be with the "+=~" object added to MSP as of max6, gen~ got me searching for it there ;p, but good to see you got your gen~ working anyways).

for mixing fft sigs, if needing a truly 'dry' signal to mix wet with, it would be inefficient to add that in pfft~(it'll be performing unnecessary phase calculations, whereas outside pfft~ it's a simple multiply).

but sorry if I don't understand why you need dry within pfft as opposed to mixed-in externally, likely I may be missing something about it...

for mixing fft sigs, if needing a truly 'dry' signal to mix wet with, it would be inefficient to add that in pfft~(it'll be performing unnecessary phase calculations, whereas outside pfft~ it's a simple multiply).
but sorry if I don't understand why you need dry within pfft as opposed to mixed-in externally, likely I may be missing something about it...


Thanks! RE the warper, I thought what I'd patched before in msp should have worked but for some reason it didn't. I still don't understand why...

So back to the topic, I've got 6 different processes going on inside the same pfft~: frequency scaling, shifting, warping, spectral gating, spectral delay, and blurring. Each one is encapsulated inside a poly~ inside the same pfft~ so I can mute/bypass them. Finally, the pfft~ itself is encapsulated inside a poly~ so that I can dynamically change the number of instances (for multichannel processing) and load difference FFT sizes.

I would like to be able to mix different levels of each process without having to do extra fft/ifft which, in this context, would be more inefficient. I'd post the patch here but it's still very much in progress!

Thanks! RE the warper, I thought what I'd patched before in msp should have worked but for some reason it didn't. I still don't understand why...  

ah, i c now, you want submixes. good idea.

(i generally put pfft~ inside poly~'s first(so that the fftin/out is also muted on the fly), and this allows for easily rerouting the chain of effects... but if you have "frequency scaling, shifting, warping, spectral gating, spectral delay, and blurring" all within a set fx-chain order(never reordered), then your design is probably most efficient since you won't be taking separate ffts for each process.

still, one thing to keep in mind: you can't use in~ or get regular signals into pfft~, can only send messages, so all of the mixing control will be updated at scheduler-rate(start of each pfft~ vector)... edit: although, i haven't tried send~/receive~ in a pfft~, not like in pvoc tutorial but receiving-from and sending-to outside the pfft~ that is... i should do that... sometime... later... ;p)

a dry mix in the frequency domain is more computation than in amplitude(and it will always be a 'resynthesis'): it involves keeping track of 'instantaneous frequency'(phase differences) using framedelta~, then adding phase differences back at the end of the chain using frameaccum~(see helpfiles), while multiplying both frequencies and amplitudes by an interpolation factor(mix).

^but that's my own very basic description of what will need to be done for each process.

i could only help by telling you to see gareth loy's "musimathics vol. ii"

(i could be wrong, but seems like it will be very complicated)

(i generally put pfft~ inside poly~'s first(so that the fftin/out is also muted on the fly), and this allows for easily rerouting the chain of effects... but if you have "frequency scaling, shifting, warping, spectral gating, spectral delay, and blurring" all within a set fx-chain order(never reordered), then your design is probably most efficient since you won't be taking separate ffts for each process.
still, one thing to keep in mind: you can't use in~ or get regular signals into pfft~, can only send messages, so all of the mixing control will be updated at scheduler-rate(start of each pfft~ vector)... edit: although, i haven't tried send~/receive~ in a pfft~, not like in pvoc tutorial but receiving-from and sending-to outside the pfft~ that is... i should do that... sometime... later... ;p)

a dry mix in the frequency domain is more computation than in amplitude(and it will always be a 'resynthesis'): it involves keeping track of 'instantaneous frequency'(phase differences) using framedelta~, then adding phase differences back at the end of the chain using frameaccum~(see helpfiles), while multiplying both frequencies and amplitudes by an interpolation factor(mix).
^but that's my own very basic description of what will need to be done for each process.
i could only help by telling you to see gareth loy's "musimathics vol. ii"
(i could be wrong, but seems like it will be very complicated)


Thanks for the info. This is stuff I didn't know. Will definitely look at the Loy text.

also eric lyon's "Designing Audio Objects for Max/MSP and PD" can help you apply the information from 'musimathics' to msp(specifically the idea of instantaneous frequency being 'implicit' in phase differences... he goes through a few ffts things near the end of the book).

Thanks for that. I've wanted to get hold of that book for ages...

BTW, with poly~ inside pfft~, you can still reorder the components of an fx-chain using selector~. This way, for each audio channel, you only do the fft/iff once for the entire fx-chain, which saves on the CPU. And with poly~ inside pfft~ inside poly~ you can mute the fft/ifft too. So you get the best of both worlds!

But I've found that gizmo~ does something strange: if placed inside the same pfft~ and after my spectral warping implementation, it somehow 'corrects' some of the warper's painstakingly achieved mess-up (specifically, the bin amplitude accumulation). Must be to do with the peak extraction and phase correction built into gizmo~. So I'm having to put gizmo~ inside a different pfft~, which is a shame. The dark secrets of FFT never fail to baffle me!

Ya definitely... I was a little scatterbrained between different threads when I wrote previously about it. Good call :)

That's weird about gizmo, I haven't used it much(heard it was expensive on CPU which always makes me want to find my own solution... But that usually means over a longer period of time, hehe). Maybe gizmo~ has an internal way of forcing the ordered output of bins in general. Good to find this out though, I should try it out more.

FFT and mixing signals