Cytomic "The Scream" stomp box distortion plugin

[acousticglue]

I have really been digging into mods and very interesting and now am brainier. This plugin is amazing and yes lighter CPU would be welcome. The when is it done question actually referred to presets as I am interested to see other settings as well

[Vertion]

Andy Cytomic, there is a new thread for you awaiting a response.

viewtopic.php?f=1&t=512525

[sleepcircle]

dude, posting the same thing in 3 topics is kind of spammy.

[Vertion]

Not if it's important to the community. Can you fathom a Cytomic synth..? Hardware synths might truly be a thing of the past if Andy get involved.

[zvenx]

Andrew has already said before that he plans to do a(nother) synth. (he did the DCAM for fXpansion)
rsp

[andy-cytomic]

Andy Cytomic, there is a new thread for you awaiting a response.

viewtopic.php?f=1&t=512525

Thanks for letting me know For the record Vertion is a good guy, he's just excited about an upcoming synth from Cytomic, so please cut him some slack!

[andy-cytomic]

How many presets will the final release have?
I have been digging in to the demo, but I'm curious what other people can do with the MOD section.

I don't want to overwhelm people with presets. I will provide around 50 presets, and each one will be a good starting point for further experimentation. On top of that customers will be able to use the randomisation engine to generate thousands of variations, so the presets will be designed with this in mind.

[andy-cytomic]

I have really been digging into mods and very interesting and now am brainier. This plugin is amazing and yes lighter CPU would be welcome. The when is it done question actually referred to presets as I am interested to see other settings as well

The lighter CPU mode has taken some time, but I understand how important it is to deliver for v1.0 and allow The Scream to be used in large projects more easily when people don't need the full detail currently on offer, or only need it for render. I've finished c++ code generation of the automated solver functions, and I'm now working on the support functions needed to make a full c++ class that will compile and run. I'm very excited to soon have some near final numbers of how much CPU it will take for a fully SSE / AVX optimised c++ version of the PDK solver, and also at the possibility of easily being able to use this same automated circuit solver to tackle other famous stomp boxes much more quickly!

[andy-cytomic]

News update: I've got the first SSE2 optimised cut down scream solver up and running using the PDK method to speed up solving of circuits using pre-computed tables. Up until now it hasn't been "real" in terms of I've only ever seen plots in maths packages and python, not properly optimised c++ code running in a plugin, so this is a huge milestone!

This has been months of R&D and months of implementation to get to this point, and it's looking very promising already, but I expect it to be more useful for more difficult circuits, this circuit just has the main op-amp voltage rail clipping and the main drive diodes, I expect more speedup as I add more non-linearities, and optimise things better.

It's producing pretty much identical results to the full brute force MNA solver (which is what The Scream currently uses), but performs x2.25 times faster (2.0% vs 4.5% [*]) on "difficult" signals like gaussian white noise. It performs slightly slower that brute force MNA on "easy" signals like a low frequency sine wave, but I've posted the results of that as well so you can see how clean and identical the spectrums are:







edit: [*] this is for the very much simplified circuit of just the op-amps voltage limiting diodes, and the main drive diodes, not the full circuit of The Scream that is in the current beta.

[sqigls]

you're the flipping coolest.

Thank you for your hard work!


bring on the Ibanez green

[sleepcircle]

nice! this must be a real rush for you, to finally get it to this point. i'm very happy for you.

[andy-cytomic]

nice! this must be a real rush for you, to finally get it to this point. i'm very happy for you.

Thanks! Yes this is brilliant and a bit of a relief since I wasn't 100% sure until I did the final optimisations and profiling that the method would indeed be fast enough and accurate enough to make it to final production. It is definitely both

[andy-cytomic]

I've done some testing against the current full MNA solver and it's coming in at around 1/4 the cpu consumption, and for harder signals it's as low as 1/10th the current cpu

This is all while still supporting switching of the diode types and count, and adjusting the bias, and being able to edit almost all of the component values (in stereo!).

[farfadetfarfelu]

Great news, my old "live" laptop is eagerly waiting for the release !

[andy-cytomic]

Great news, my old "live" laptop is eagerly waiting for the release !

Yes, the final release will be much friendly in terms of CPU with the new MD detail level (currently it's HD only). I've just completed what is pretty much near final DSP for the MD mode, enough to give a very accurate estimate of what to expect.

The speedup will be around x3 to x8 compared to the current solver

Easy signals 44.1kz: 2.7% MD, 8.5% HD
Hard signals 44.1khz: 3.3% MD, 26% HD

And the best thing is you can't hear any difference between two apart from some edge cases, and you still get to edit almost every single component value in the whole circuit, including the main drive diodes and all the filtering capacitor values

The last 6 months or so of hard slog and R&D have definitely paid off since the solving methods apply to all circuits, not just The Scream. This is the culmination of around 6 years of solid work on automated circuit solving for audio processing, and boy does it feel good!

Time to tidy up this latest round of DSP, then it's all full steam ahead on getting The Scream finished for the final release.