Repro-1 (out now)

[beely]

Just had a look at the poll results on the GS thread - interesting how the poll results differ.

Basically, it seems to suggest the more experience in the typical GS forum user would discount filter 1 (the cheapest) almost immediately, whereas 1 scored much more highly here, probably with a higher spread of users more used to software only and less of boutique synths (I generalise of course).

There was some quite cogent analyses of the filters over there, as well. Props to them...

And the maths is *way* over my head... heh!

[urosh]

Just had a look at the poll results on the GS thread - interesting how the poll results differ.

Basically, it seems to suggest the more experience in the typical GS forum user would discount filter 1 (the cheapest) almost immediately, whereas 1 scored much more highly here, probably with a higher spread of users more used to software only and less of boutique synths (I generalise of course).

Actually, #5 is "cheapest", and it's tied for 1st place with #4 at GS poll.

[fmr]

Just had a look at the poll results on the GS thread - interesting how the poll results differ.

Basically, it seems to suggest the more experience in the typical GS forum user would discount filter 1 (the cheapest) almost immediately, whereas 1 scored much more highly here, probably with a higher spread of users more used to software only and less of boutique synths (I generalise of course).

Actually, #5 is "cheapest", and it's tied for 1st place with #4 at GS poll.

For what I read, the #1 is also "cheap" (and it scored second). Anyway, Urs said that in the end none of them is really that "cheap".

[urosh]

Urs, I have few questions:
do you use in practice 0.000001 error limit for iterations (you mentioned that number on blog)? If you do, have you tried and compared with limit at -180dBFS (as opposed to -120dBFS)?
And, what was highest number of iterations you've got in practice(let's say, with high level white noise input, or something like that), if you log that kind of thing? I guess 4-6 you mentioned elsewhere are averages, if 6 is absolute highest than screw filter solver it self, predictor you use is made by help of black magic.

BTW, I've posted example in u-he forum some days ago, of situation where even sanity checker for 1 and 4 failed. Pivotal is great, but I'm not sure that even with smooth tanh it would absolutely always produce sane result fro abused filter.

[beely]

Actually, #5 is "cheapest", and it's tied for 1st place with #4 at GS poll.

Fair enough. Though the point remains - the main difference in voting results was that 1 *sounded* cheapest, and got votes here while didn't really at GS (though there weren't that many votes there in total, so it may not be that representative). Otherwise the voting was similar.

Just thought it was interesting...

[Urs]

Urs, I have few questions:
do you use in practice 0.000001 error limit for iterations (you mentioned that number on blog)? If you do, have you tried and compared with limit at -180dBFS (as opposed to -120dBFS)?
And, what was highest number of iterations you've got in practice(let's say, with high level white noise input, or something like that), if you log that kind of thing? I guess 4-6 you mentioned elsewhere are averages, if 6 is absolute highest than screw filter solver it self, predictor you use is made by help of black magic.

I think I'm using a typical Work-In-Progress formula: 0.01 / 65536.0

The number of iterations/CPU depends on settings (input gain!) but also on accuracy of computation. Sometimes using a "real" division is faster than using a reciproce estimate. Sometimes using an improved initial guess is faster than not. In the end we'll measure the actual CPU consumption and pair certain optimisations with certain settings.

Furthermore I think that maybe the delta-method described by Andy Simper may yield higher numerical accuracy and thus fewer iterations.

For the final conclusion on average number of iterations we'll use the factory presets and measure how long they take to render a few MIDI files. We have a test rig for that.

BTW, I've posted example in u-he forum some days ago, of situation where even sanity checker for 1 and 4 failed. Pivotal is great, but I'm not sure that even with smooth tanh it would absolutely always produce sane result fro abused filter.

I'm not sure either. Thing is, the original method posted by Mystran uses the s-values for the tanh()-terms - which in bilinear integration is half a sample delayed from the next inputs. This doesn't work in our particular case because the tanh() (or whatever waveshaper) isn't taken from the input voltages separately, it's taken from their sums. So the dynamic may be a bit more complicated.

[foosnark]

3 is my least favorite

I guess this is confirmation that I am secretly a robot, unconcerned with squishy meatbag preferences for analog over digital.



After all, I did trade one fully analog and one hybrid analog desktop synth for one digital and one ludicrously digital VA synth, and I regularly run my Microbrute through various FFT effects and sample rate reducers.

...but this has been a cool experiment and I'm enjoying reading the results.

[wagtunes]

Where in this thread has it actually been revealed which filter is cheapest, most expensive and so on? I can't find any such information.

[beely]

Where in this thread has it actually been revealed which filter is cheapest, most expensive and so on? I can't find any such information.

In the link Urs posted a few pages back, which reveals the results and his conclusions.

viewtopic.php?f=1&t=460687&start=765#p6453653

Or you could just go back to the first post, which has been updated with the results link.

[suthnear]

This plug in sounds really great, Urs, maybe the best analogue model yet. The envelope shape and behaviour, particularly, is spot on. I'd love to see this make its way into Diva (where the sound breaks up a bit with very short decay for either the filter or the amp).

Fwiw, 1 produced some really nice non-analogue artefacts at high resonance that might be worth exploring in the context of Z3 (d'oh, I see you already mentioned this in your PDF discussion).

p.s. you almost certainly know this, but there's a bug with preset saving whereby it seems to get stuck with envelope->VCA, regardless of what the switch is set to

[mholloway]

So, Short Version: #3 is best in terms of functionality.

I can live with that!

Very interesting read, if often a few notches above my pay-grade

[db3]

All very interesting. Although I suspect most people selected the option which simply sounded the best to them...and the results would have been different had it been a blind poll.

[aMUSEd]

I thought that was the point, to select what people thought sounded best to them? It was blind in that no one knew which one was actually the 'best one' in terms of design. People could see each others votes I suppose but even so very few people actually specifically stated 3 was the best in the thread and some of us voted for 3 without even reading the thread (I know I did and 3 was easily the one that stood out for me).

[db3]

The purpose was to select the most analogue sounding option, not necessarily the best sounding option. A blind poll is one in which voters cannot see how others have voted. People will naturally be swayed by how others vote, and from comments in the thread.

[aMUSEd]

Yes but I think it's never been clear what the difference is between 'most analog' sounding and 'best sounding' - in practice I think people would have just used their ears and picked what they felt sounded most analog assuming that would also be what they thought sounded the best. For me 3 sounded the smoothest and most squelchy and sounded best swept at high resonance so I voted for it, didn't care what others thought, that wasn't a factor (in fact I voted on that before reading this thread and anyway most people in the thread stated they didn't like 3 at first anyway).