Cytomic "The Scream" stomp box distortion plugin

[andy-cytomic]

Hey Andy, I noticed another typo or 2 in the User Manual:

"An often sited source of the classic tone of the original Tube Screamer pedals is the use of a
particular mass market op-amp chip, the JRC4558D" (page 9)

"Sited" should be "cited", like a citation.

The potential 2nd typo is the previous page (8) says this:
"but we have not sighted the original patent"

This may or may not be the same mistake, with a slightly different misspelling. Or, an argument could be made that the meaning here is that you have not put your eyes on the original patent, though this would be an awkward turn of phrase, rather than simply saying, "but we have not seen the original patent."

Thanks for spotting those, seen is definitely better. I've changed the other sentence about the op-amp chip, so now I think the paragraph scans better.

[martinjuenke]

▒▒░░▒▒▓▓▒▒░░▓▓▒▒░░▒▒▒▒▒▒░░░░▓▓░░▒▒░░░░░░▓▓▒▒░░▒▒▒▒░░▒▒▒▒▓▓▒▒

YES!

[andy-cytomic]

News: The Scream v1.2.9 is available for download.

The Scream v1.2.9 (9 Dec 2024)
https://cytomic.com/product/scream
• Fix: HD solver self noise on silent input due to numerical issue now fixed
• Fix: MD solver now more accurate to HD sound with improved solve method

[dayjob]

Stoked to have these updates. Thanks Andy.

[andy-cytomic]

Stoked to have these updates. Thanks Andy.

You're welcome! They are the practical output of a little more R&D into the more optimised and stable solutions to arbitrary circuits. I've been chatting with Dillon Sharlet, who wrote the windows only LiveSpice solver, about different strategies for this sort of symbolic circuit solving, and we've both been able to help one another clarify what works best, and how to improve our own work. Hopefully there will be some new releases of LiveSpice soon with smooth parameter changes and more efficient cpu usage of his software thanks to our discussions. Through our discussions I've been able to re-structure slightly how I form the Matrix equations a little, which has given rise to a much more elegant initial output of the symbolic solution from the solver, which means the CAS (computer algebra system) has a much easier time optimising the solution afterwards.

I'll release v1.2.10 later on today, but from v1.2.0 there is an around 60% improvement of cpu usage solving the same circuit (HD mode with full drive and loud white noise as input). Without even applying the CAS (which takes around 1 hour to fully process) I can now generate efficient solutions for internal testing, which will help reduce the turnaround time trying out different models and getting a rough idea of how efficient they will be for production use.

This is all really getting things in place for the next round of products, it's easy to test out these optimised approaches on an existing circuit before applying it to new ones.

[jtsterays]

Jesus just as I though you have reached the limit of The Scream optimization. You've set a new benchmark for CPU usage to emulation detail ratio, quite mind blowing. Hopefully you will tackle more complex gear like the Culture Vulture in the future.

[andy-cytomic]

Jesus just as I though you have reached the limit of The Scream optimization. You've set a new benchmark for CPU usage to emulation detail ratio, quite mind blowing. Hopefully you will tackle more complex gear like the Culture Vulture in the future.

Thanks! I think I'm very close to the actual limit of optimisations possible now while still allowing for highly parametric and highly accurate non-linear realtime circuit emulation - where things like the power voltage as well as the regular controls are all parameters, and every component's model parameters can be randomised to generate variations. It's a tough problem in general, and I am very happy with the efficiency vs detail I'm able to deliver. Also the generated code is amazingly lean and concise, and dare I say elegant for something that is automatically generated by an algorithm.

Tube modelling is definitely on the way, which I'm pretty confident I'll get excellent results. I also want to tackle detailed and accurate transformer modelling, but I'm less confident, but should still be able to get something that sounds good at least.

[andy-cytomic]

News: The Scream v1.2.10 is available for download.

The Scream v1.2.10 (10 Dec 2024)
https://cytomic.com/product/scream
• Fix: Diode count of 2 and above now correctly applied to the solver, in the last few builds a value of above 1 was set to 1, which would apply too much drive to the signal

[jtsterays]

Jesus just as I though you have reached the limit of The Scream optimization. You've set a new benchmark for CPU usage to emulation detail ratio, quite mind blowing. Hopefully you will tackle more complex gear like the Culture Vulture in the future.

Thanks! I think I'm very close to the actual limit of optimisations possible now while still allowing for highly parametric and highly accurate non-linear realtime circuit emulation - where things like the power voltage as well as the regular controls are all parameters, and every component's model parameters can be randomised to generate variations. It's a tough problem in general, and I am very happy with the efficiency vs detail I'm able to deliver. Also the generated code is amazingly lean and concise, and dare I say elegant for something that is automatically generated by an algorithm.

Tube modelling is definitely on the way, which I'm pretty confident I'll get excellent results. I also want to tackle detailed and accurate transformer modelling, but I'm less confident, but should still be able to get something that sounds good at least.

With your framework now really matured, can we expect faster product output from you? You mentioned it would take 2 months for a pedal, but what about compressors and other stuff? Hopefully we wont have to wait until the middle of next year for Glue V2 or a decade for a synth (Serum 2 has been in development for 5 years now, and that's with 3 people). And why is transformer modelling harder? If even you're not confident about transformers then I can assume the rest of the market haven't figured it out yet, which is quite disappointing.

[Funkybot's Evil Twin]

I also want to tackle detailed and accurate transformer modelling, but I'm less confident, but should still be able to get something that sounds good at least.

It definitely sounds tricky to get to "cytomic-level accuracy" on transformers. I suspect most companies' transformer "models" are just combinations of basic saturators and filters designed to approximate the impact of the transformer on the signal's output. In contrast, I believe you're talking about creating a full-blown digital model of the transformer itself. At which point, you're talking about modeling the impact of different transformer core materials, different dimensions/geometry, different number of primary/secondary windings, different bell cover materials, etc. That sounds like a pretty bonkers task and one that would be computationally expensive by itself. I wish you luck, but that also seems like it could be a months or years-long task in and of itself. So at least bang out some more simple guitar pedals in the meantime please.

[jtsterays]

I also want to tackle detailed and accurate transformer modelling, but I'm less confident, but should still be able to get something that sounds good at least.

It definitely sounds tricky to get to "cytomic-level accuracy" on transformers. I suspect most companies' transformer "models" are just combinations of basic saturators and filters designed to approximate the impact of the transformer on the signal's output. In contrast, I believe you're talking about creating a full-blown digital model of the transformer itself. At which point, you're talking about modeling the impact of different transformer core materials, different dimensions/geometry, different number of primary/secondary windings, different bell cover materials, etc. That sounds like a pretty bonkers task and one that would be computationally expensive by itself. I wish you luck, but that also seems like it could be a months or years-long task in and of itself. So at least bang out some more simple guitar pedals in the meantime please.

DIYRE SSL, the hardware Andy is using for the Glue V2 includes a transformer drive knob, so he probably will tackle transformers first.

[DNAudio]

News: The Scream v1.2.10 is available for download.

The Scream v1.2.10 (10 Dec 2024)
https://cytomic.com/product/scream
• Fix: Diode count of 2 and above now correctly applied to the solver, in the last few builds a value of above 1 was set to 1, which would apply too much drive to the signal

Does this version fallback to software rendering because of Juce 8? If not, it'd be great if you could add a configuration file that allows forcing software rendering. Every Juce 8 plugin is a black screen on my system

[andy-cytomic]

News: The Scream v1.2.10 is available for download.

The Scream v1.2.10 (10 Dec 2024)
https://cytomic.com/product/scream
• Fix: Diode count of 2 and above now correctly applied to the solver, in the last few builds a value of above 1 was set to 1, which would apply too much drive to the signal

Does this version fallback to software rendering because of Juce 8? If not, it'd be great if you could add a configuration file that allows forcing software rendering. Every Juce 8 plugin is a black screen on my system

Yes, this is a Juce 8 plugin. I fallback to the software render on Windows only. Can you please email support@cytomic.com with a bug report?

[andy-cytomic]

I also want to tackle detailed and accurate transformer modelling, but I'm less confident, but should still be able to get something that sounds good at least.

It definitely sounds tricky to get to "cytomic-level accuracy" on transformers. I suspect most companies' transformer "models" are just combinations of basic saturators and filters designed to approximate the impact of the transformer on the signal's output. In contrast, I believe you're talking about creating a full-blown digital model of the transformer itself. At which point, you're talking about modeling the impact of different transformer core materials, different dimensions/geometry, different number of primary/secondary windings, different bell cover materials, etc. That sounds like a pretty bonkers task and one that would be computationally expensive by itself. I wish you luck, but that also seems like it could be a months or years-long task in and of itself. So at least bang out some more simple guitar pedals in the meantime please.

Ok, I perhaps take it back about accurate non-linear transformers, but there is still a little R&D to implement them in a low cpu and stable manner, but I can handle that

In papers I read a while back it looked like I'd have to model the spatial side of things to get the core saturation and hysteresis correct. I just did another search and found this 2009 paper which looks very promising: https://ieeexplore.ieee.org/document/4802905 . They look like they get decent results just using a gyrator, and a specific type of non-linear capacitor and non-linear resistor in series "inside" the gyrator section. I've already implemented non-linear resistors (eg diode) and non-linear capacitors in The Scream, so all this is very promising. A gyrator is a trivial linear element I already support, and it transforms a capacitor into doing the same thing an inductor does in a circuit through a transformation (power conjugate pairs): https://en.wikipedia.org/wiki/Gyrator–capacitor_model

This is important for tube amplifiers that are coupled to speakers via a non-linear transformer. The paper linked below uses the above paper to model the non-linear transformer for this purpose:
https://asp-eurasipjournals.springerope ... 011/347645

[andy-cytomic]

News: The Scream v1.2.11 is available for download.

The Scream v1.2.11 (11 Dec 2024)
https://cytomic.com/product/scream
• Fix: diode count can now reach below 1, so the combo box for choosing the diode count now works with 0.5
• Fix: switched back to software renderer on macOS because of reported black screen UIs
• Fix: improved handling of limits for various component values reaching 0 or generating large currents