Hello!

I was wondering what method all those new additive synths use for creating those sine wave partials. Are they wavetables or are they something like FFT reversed? If they are created independently I think it is possible to not play those partials that go above Nyquist freq. Are anyone tested Cameleon, Vertigo, Cube... in spectrum analyzer? If so do you find aliasing or quantization distortion etc. ? Additive synthesis is a way to make aliasing free sounds if the synth can create pure sine wave partials. I hate aliasing in pad and melody sounds!

Hi Ville,

There is no aliasing in Cameleon - as the frequency of harmonics approaches nyquist, they are faded out. I think this was commented on as an advantage of Cameleon in the Keyboard review (where it got a Key Buy award ).

You can get more info, and download the demo at:
http://www.camelaudio.com/products.php?pID=28

Ben

but what if you combine two slightly out of phase sines with frequencies above Nyquist? this should give you intermodulation in the audible range although the two sines are inaudible, so in a sense they indirectly contribute to the sound, and hence should be preserved?

Thanks Ben!

Now every developer whos additive synth is alias free should inform me about it because now Cameleon is my first choise if I buy an additive synth!
I'm wondering why developers don't mention it if their synth is alias free!

i thought aliasing was more a matter of degree than yes/no

ville wrote:

Thanks Ben! Now every developer whos additive synth is alias free should inform me about it because now Cameleon is my first choise if I buy an additive synth! I'm wondering why developers don't mention it if their synth is alias free!

and don't forget to offer a lamb to the pageant gods of anti-aliasing

monsterbeetle wrote:

but what if you combine two slightly out of phase sines with frequencies above Nyquist? this should give you intermodulation in the audible range although the two sines are inaudible, so in a sense they indirectly contribute to the sound, and hence should be preserved?

No, for the intermodulation tone to appear, a nonlinear process must occur. Simply adding two tones does not generate any combination tones.

ville wrote:

Thanks Ben! Now every developer whos additive synth is alias free should inform me about it because now Cameleon is my first choise if I buy an additive synth! I'm wondering why developers don't mention it if their synth is alias free!

Alias free? Because most aren't alias free. It's not like it's a 'door' you can simply shut, it's a steep low pass filter. Then, it makes a difference what sample rate you use on top of that.

Still, I wouldn't buy a synth based on if it aliases or not, I'd buy it because I like the sound. There's a lot of great synths out there that alias that have an awesome sound, Vanguard being one of them.

Devon

at last I can see the convergence of 'sounds like virus' and 'anti-aliasing' thread

matter and anti-matter within the same vessel

wrench45us wrote:

at last I can see the convergence of 'sounds like virus' and 'anti-aliasing' thread matter and anti-matter within the same vessel

it's not going to take long before aliasing becomes a religion.. and we all know how useful those are

Svante wrote:

monsterbeetle wrote: but what if you combine two slightly out of phase sines with frequencies above Nyquist? this should give you intermodulation in the audible range although the two sines are inaudible, so in a sense they indirectly contribute to the sound, and hence should be preserved? No, for the intermodulation tone to appear, a nonlinear process must occur. Simply adding two tones does not generate any combination tones.

Well if you actually play two sounds, your ears can contribute the non-linear function.

Hi DevonB!

In additive synthesis you build the sound from sine wave partials. If the partials are wavetables of mathematically created sine waves each created for particular frequency and not pitch shifted there should be no aliasing at all. You dont need low pass filter because you simply don't play sine waves above Nyquist.
I have made experiments with resynthesis in Bidule and now in Tracktion. I connected 16 sfz-samplers to complex setup and those samplers play sine wave partials which I created in coldwave with expression evaluator. I made all required partials for every note from 20Hz to 20000Hz. I simply don't made samples above Nyquist. My little experiment worked like a charm. It's like a combination of vocoding and 16 partial additive synth. In the future even if it sounds little stupid somebody could make GUI to synth which is like analogue synth but the engine is additive synth in disguise!
Freedom from aliasing is not a religion or is it!? I simply don't want any unwanted digital artifacts.

Yes, that's a perfect world example. Ben just said above that he fades the sound out as it reaches Nyquist. If there were no alias components in the additive, there would be no need, but obviously there is some need if he did indeed use a filter.

Really, if you want to avoid aliasing altogether, record from pure analog synths to analog tape. Problem solved.

Still, I think the whole aliasing bit is overhyped and misunderstood. I do not consider aliasing to be bad or good, just a 'feature' you either like, or you don't. If it was so hated, the DX7 never would have sold in the quantities that it did.

Devon

Well as a matter of fact I often record my analog synth and my circuit bent old analog yamaha organ! And when I make samples from them I sample every key! I guess I'm a little bit fanatic about these things! If your AD-converter is normal you don't need tape recorder to avoid aliasing because there is an analog filter before converter.
When it comes to drums and pecussion I'm not so fanatic about aliasing and sometimes it even sounds cool on drums!

Quote:

You dont need low pass filter because you simply don't play sine waves above Nyquist.

Absolutely.

Quote:

Yes, that's a perfect world example. Ben just said above that he fades the sound out as it reaches Nyquist. If there were no alias components in the additive, there would be no need, but obviously there is some need if he did indeed use a filter.

I chose to fade the partials out (simply by scaling their amplitudes - no LPFs needed), because I believe its better to do this, than simply have them disappear when above nyquist. This is all fairly academic, as its only in the last few kHz before niquist that I fade them. Its only really if you run the synth at 22kHz or below that it makes an obvious audible difference. In this case I prefered the upper partials to fade out smoothly, rather than have then just disappear when they go over nyquist (ie. 11kHz in this example), which sounds a bit odd. This is all getting rather perfectionist (anal - some might say) - but hey, I think its worth it

The important point is that with additive, its possible to create alias free waves easily.

Ben

DevonB wrote:

Really, if you want to avoid aliasing altogether, record from pure analog synths to analog tape. Problem solved.

Actually, an analog tape deck can potentially exhibit its own form of aliasing, if you try to record a very strong signal that approaches the frequency of the tape deck's bias oscillator.


later,
McLilith

if you want to completelly avoid aliasing, don't record, only play it live. avoid digital rack effects, only use analog pedals.

protools and nuendo users don't know what's aliasing, cubase, logic and samplitude users ocassionaly suffer from it, the rest of mere mortals suffer from it badly

:::::::::: flames here :::::::::::::::::::::

Ben [Camel Audio] wrote:

The important point is that with additive, its possible to create alias free waves easily. Ben

Ok, I gotcha. When you have control over individual partials, avoiding aliasing is a lot easier. Makes sense. I just wasn't thinking of it in that aspect. IE a big DUH on my part!

Devon

All my synths do this as well (Doppelmangler, WNAdditive) and should be alias free for the most part.

You can also choose synths that allow you to render at higher settings.

Robert

I just studied 4 of the 5 currently available additive synths. There is no aliasing, and if so, not in a frequency I can hear in the two mentioned so far, Camel Audio Cameleon and WhiteNoise Additive. I can assue you similar results in VirSyn CUBE or the additive oscillator in TERA. As for ConcreteFX there may be some aliasing. I don't have a scope and to my ears it is either a deliberate choice with Adder or possible aliasing. If so, I like it's effect.

The area that all the additives really deserve study is on how they implement additive synthesis. For example:
Cameleon - interpolated sample data with 4 oscillators
Additive - interpolated sample and visual data along with traditional partial editing tools
CUBE - all sine waves before they hit any other part of the synth. Partial and spectra control as well. It's new sample resynthesis addition sound exceptionally clean.
Adder - my have aliasing, if it does it is well designed in the synthesizer's sound. Sample data is interpolated into 32 partials that act like individual oscillaors. Big deal, all additves work that way. Wrong, they often collect higher frequencies into a partial. In Adder all 32 have the entire sound engine at their disposal. This is different as if one wanted to they could use all the various, and there are a lot of sound shaping tools for Adder, which means that their algorithm must be substantially different than the others.

None of these synths work alike and yet they produce unique sound, some of the highest quality you'll find in VST instruments.

x_bruce wrote:

The area that all the additives really deserve study is on how they implement additive synthesis. For example: Cameleon - interpolated sample data with 4 oscillators...

Erm, not exactly. There is only one "oscillator" in the senes that you are using the term. However, there are 4 sets of harmonic and noise profiles (voice profiles) which that single "oscillator" can blend in variable amounts, to create an instrument.

I'm also not sure what you mean by "interpolated sample data". I hope you realize that Cameleon stores no samples in its presets. What gets stored is detailed references to how loud, and exactly what pitch, a particular sine-wave partial should be at any particular point in a note's timeline.

In a related manner, there is also the equivalent of a constant noise generator connected to a multi-band equalizer. (I'm talking about lots of distinct frequency bands in this particular EQ.) The voice presets also store references about how loud each frequency band of noise should be at any particular point during the progression of a note.

Of course, there is a lot more to this synth, but that is a very basic description of the generator section. Cameleon offers all sorts of options to further enhance the raw sounds generated by this section of the synth.

In summation, and to paraphrase "Spoon Boy" -- "Don't try to avoid aliasing in the sample playback, for that would be impossible. The thing to realize is there are no samples."


later,
McLilith

monsterbeetle wrote:

but what if you combine two slightly out of phase sines with frequencies above Nyquist? this should give you intermodulation in the audible range although the two sines are inaudible, so in a sense they indirectly contribute to the sound, and hence should be preserved?

Wouldn't you just add the frequency you can hear as another partial?

Quote:

In this case I prefered the upper partials to fade out smoothly, rather than have then just disappear when they go over nyquist (ie. 11kHz in this example), which sounds a bit odd

Isn't this how it should be? I don't find it that odd

For ex, when you downsample something, you wish your filter had no slope at all. Here you can really have no slope at all, so why choose one?

That way a partial doesn't suddenly disappear as it's pitch is rising. This is probably the right way to do things but in practice I've found that it doesn't seem to make much difference.

in practice i wouldn't think you'd want to have much energy that close to the nyquist, anyway -- most "real world" samples will naturally be attenuated close to 22kHz by the sampling filter (if not already due to natural processes).

but doesn't your ear already attenuates them? starting at 16khz it should start to attenuate, so no need to attenuate the signal..

well maybe it matters if your samplerate is below 44khz

oh yeah, there's no way i could hear 20kHz -- but i might get an unpleasant surprise if i put that signal through a non-linear process. a surprise, because a similar sound sampled from the real world might not have enough energy up there to alias noticeably.

anyway, from what whitenoise says, it sounds like it isn't really an issue - so i guess i should have said "in theory" instead of "in practice"

i will tell you, though, that i get all kinds of annoyed with my morpheus for having a lot of energy in the 16+ kHz band. you don't really hear this much if you just use it normally, but send it into a feedback loop on my se-50 and the aliasing quickly becomes unbearable.

Quote:

For ex, when you downsample something, you wish your filter had no slope at all. Here you can really have no slope at all, so why choose one?

I initially thought that, but there are reasons ...

Quote:

That way a partial doesn't suddenly disappear as it's pitch is rising. This is probably the right way to do things but in practice I've found that it doesn't seem to make much difference.

Yes - absolutely - it doesn't make any difference when you're sampling at 44kHz or greater. (I can't hear much above 17kHz). I did prefer this approach for 22kHz though, for the reason Dave mentioned. I figured that if this is what I prefered in the audible range, it would probably be a good idea to do a similar thing for higher sampling rates too. Also as autloc mentioned, there are cases where you're processing the signal afterwards, where its good to have this smoothness.

Anyway, its not a big deal at all. I'm just replying to explain why I made the decisions I did.

Ben

Huuuuuuuuuuuuhuuuuu ! Another thread about aliasing and I was missing ?
This is the 0xFF85A43c 'd thread about aliasing, and people spit when they hear about it. After all, i think if you can't notice aliasing in sound than it's ok, but ; Imagine you feed a aliased synth to a distortion/overdrive effect. Nasty sound, just try it. It's hardly possible to make alias free sound, when your bandwidth is only 22050 Hz. If you use a higher sample rate than it's better.
My wavetable oscillator is using different wavetable for each MIDI note, so I have a 2 MB wavetable, but at least it does not alias. There are some discrete frequencyes when aliasing occures, but their level is not significant, and it's from 18kHz upto 22050 Hz.

Quote:

It's hardly possible to make alias free sound, when your bandwidth is only 22050 Hz

this thread is about additive synthesis, of course it's possible & easy. It's just a matter of not playing the sines above nyquist

Ben [Camel Audio] wrote:

Hi Ville, There is no aliasing in Cameleon - as the frequency of harmonics approaches nyquist, they are faded out. I think this was commented on as an advantage of Cameleon in the Keyboard review (where it got a Key Buy award ). You can get more info, and download the demo at: http://www.camelaudio.com/products.php?pID=28 Ben

What Ben Said!!

Cameleon is da bomb! I just tried the demo! Awesome sounds and creative potential...

mauseoleum wrote:

....nuendo users don't know what's aliasing..

Quote:

...cubase... users ocassionaly suffer from it

Eh? Seems you know something about the cubase/nuendo engine that I don't. Please explain.

Thanks