Complex waveforms: calculated or samples

[Essent]

Most soft-synths nowadays have, apart from the basic waveforms saw / pulse / triangle (which are easy to compute), more complex waveforms. I wonder if those complex waveforms are computed or if it are just samples.

[jupiter8]

Mostly samples but i'd say that's a distinction without a difference. It's what you can do with the waveforms that matters.

[Deisss]

Most soft-synths nowadays have, apart from the basic waveforms saw / pulse / triangle (which are easy to compute), more complex waveforms. I wonder if those complex waveforms are computed or if it are just samples.

Likely wrong forum as there is one for development.

Short answer: anything can be used. It can range from samples to epicycles/formulae, with in between some sort of combination of it (interpolation and so on). Not to mention some attempt to model electrical circuit (as code)...

Everything is possible, each results got tradeoff to accept, so it's a matter of what you want to achieve.

[Ah_Dziz]

Very often an array being read by a phasor (so pretty much just a sample).

[Essent]

Interesting stuff. So when a company like Arturia claims they recreated a Juno / Jupiter, did they filled their synth with samples? Cause one sample probably won't do. Or would they have some algorithm to stretch a sample?

[jupiter8]

Interesting stuff. So when a company like Arturia claims they recreated a Juno / Jupiter, did they filled their synth with samples?

While it's certainly possible i seriously doubt that. The thing is you need extra algorithms for pitch shifting and anti aliasing and what nots. In the end it doesn't really matter as long as the result is good. The way you store the information is such a minor part of the whole it really doesn't matter. That's why i said it's a distinction without a difference.

[Essent]

Interesting stuff. So when a company like Arturia claims they recreated a Juno / Jupiter, did they filled their synth with samples?

While it's certainly possible i seriously doubt that. The thing is you need extra algorithms for pitch shifting and anti aliasing and what nots. In the end it doesn't really matter as long as the result is good. The way you store the information is such a minor part of the whole it really doesn't matter. That's why i said it's a distinction without a difference.

OK, so when are they calculated, when are it samples? Say a Moog saw (which is not my definition of a saw), is it calculated or a sample?

[jupiter8]

Say a Moog saw (which is not my definition of a saw), is it calculated or a sample?

A real Moog saw is analog so it's neither. Though i'd say closer to calculates than sampled.
Something like:

Code: Select all

phase=-1.0;
phaseIncrement= frequency * something/samplerate;
phase += phaseIncrement;
if (phase > 1.0) phase -= 2.0;
sawOut = phase;
pulseOut = -1.0;
if (phase > threshold) pulseOut = 1.0;

That's the absolute basic but you'd need a lot more stuff for antialiasing and sync and stuff.
That's probably how i would start, so calculated. But if i can get the exact same result doing it another way, does it matter?

[Essent]

I am just interested in the differences and arguments. The question if it matters is for me not relevant.

[jupiter8]

When you said complex waveforms i was thinking more U-He Zebra, Vital and the like.
Zebra2 is based on "samples" but converts it to the frequency domain (FFT) "does stuff" and converts it back to "samples" so you can do all kinds of weird stuff. I'm pretty sure Vital does that too. Zebra3 otoh is all formulas as i understand it.
So you can manipulate the data in all kinds of ways before the final result.

[jupiter8]

There's also component modelling in which you have digital models of the components and connect them together. That would be calculated waveforms, no samples.

[Essent]

thx, as said, I find this al very interesting. I am not a big fan of U-He's approach, but that is just my personal opinion.

[jupiter8]

thx, as said, I find this al very interesting. I am not a big fan of U-He's approach, but that is just my personal opinion.

Which one? He does all of them.

[Essent]

When a synth starts with samples and converts it to the freq domain, you can just store the values of your sinussen and do not need the sample at all when distributing the synth.

[Essent]

thx, as said, I find this al very interesting. I am not a big fan of U-He's approach, but that is just my personal opinion.

Which one? He does all of them.

Modeling every resistor, diode, capacitor, etc.