The No.1 Website for Pro Audio
 All  This Thread  Reviews  Gear Database  Synths for sale     Latest  Trending
VA for Electronic Gearslutz
Old 1 week ago
  #1
Lives for gear
 
syntonica's Avatar
VA for Electronic Gearslutz

With the release of the new Rolands, I've noticed there's a lot of misunderstanding about what VA is, how it works, terminology, etc. So, I hope to explain it well enough that we can all at least argue with a common understanding of the terminology.

Analog Modeling: this is just fancy talk for a very complex, CPU-expensive algorithm that can take an input and does give an output. It looks at each component of a circuit for its behavior and then describes it mathematically. By combining all these equations, you get a full model.

Sample: an unfortunately very loaded term. It can mean:
1. An input captured by a microphone, digitized, and stored.
2. Any data file that contains a digitized sound, regardless of its source. Some samples can be generated by the CPU according to an algorithm.
3. A single data value amongst a stream of data values given by 1 or 2 above.
Every time I use this term, I will follow it by its number.

Oscillator: in the digital world, it's a stream of samples(3) from a sample(2) that describe digitally the shape of the wave normally created in circuitry.

Band-limited: an oscillator is band-limited when all frequencies higher than the audible range are removed. This can be done by building the wave up from so many generated sine waves, or by building bleps through an algorithm that removes unwanted frequencies from a pure waveform. There are other ways, but I just give these as common examples.

Interpolation: this is used when transposing samples for playback. There are a number of methods, from the worst--linear to one of the best--sinc. Linear just takes two samples and finds a mid-point. Sinc takes several samples to find a curve to match and then finds the needed point on that curve. Sinc sounds extremely good, but can be very CPU expensive.

There are basically only two parts of a synth that need modeling: the oscillators and the filters. All the rest can be modeled, but they are not as important to the overall sound, so I am shamelessly ignoring them.

Back to oscillators: there are two ways of generating them.

One, read a sample(3) from a sample(2) and interpolate as needed if you don't have a sample(2) at the correct frequency. The CPU cost is purely dependent upon the interpolation method used. For single-cycle waveforms, it's very cheap, in both CPU and memory, to just pre-calculate each frequency needed and put them into a lookup table. Then, no interpolation is necessary.

Two: calculate the sample(3) on-the-fly. Depending on the algorithm used, this can be quite expensive in terms of CPU. Pre-calculated lookup tables for sin, cos, tan, atan, etc. can alleviate a lot of that pain.

Back to filters: filters must necessarily be calculated on-the-fly since they do not know what their input will ever be. So, there are dozens of filter models out there. Some are purely mathematical in their creation and are very CPU-cheap to use. Modeled filters tend to be very CPU-expensive. Diva is the spendy one. Arturia filters are the cheaper ones, probably by simplifying the algorithms for speed; however, the edge cases (extreme values) may not sound as good as U-he's models.

Finally, there are anti-aliasing filters that can be applied, all of varying efficacy vs. CPU-expense. These are usually applied to an oversampled signal at the very end before the data goes out to the effects and can be used in addition to or in lieu of band-limited oscillators.

I hope this helps. Don't be fooled by marketing buzzword hype! Basically, VAs, like VSTis, are all 1s and 0s and are either read from a pre-calculated table or they are generated on the fly.

Let me know if you have any questions, or see any egregious errors.

Last edited by syntonica; 1 week ago at 12:47 AM..
Old 1 week ago
  #2
Lives for gear
 
BobTheDog's Avatar
 

Good post.

Not really an error but it might be a good idea to edit the Band-Limited section to say the frequency may well change depending on the Nyquist Frequency rather than ~20Khz.

Also you could add something about the following which always seems to cause confusion here:

DSP - Digital Signal Processing : Using digital processors to process digital signals.

The same digital Signal Processing algorithms can be run on:

CPU - Central Processing Unit : A general purpose digital processor.

DSP - Digital Signal Processor : A specialised type of digital processor optimised for Digital Signal Processing, usually with low power usage and running at low temperatures.

FPGA - Field Programable Gate Array : An Integrated Circuit that can be configured after manufacturing. These usually contain an array of logic blocks, memory and DSP slices. FPGAs can be even be programmed to contain CPUs and DSPs.
Old 1 week ago
  #4
Lives for gear
 
xanderbeanz's Avatar
VA sucks, it never got the smell of Analogue right.

Old 1 week ago
  #5
Lives for gear
 
chaocrator's Avatar
Quote:
Originally Posted by xanderbeanz View Post
VA sucks, it never got the smell of Analogue right.
VA rocks, it packs a lot of features in a small/compact box.
i can sacrifice that smell of analogue to be able to bring my whole rig in a backpack to the gig.
Old 1 week ago
  #6
Lives for gear
 
xanderbeanz's Avatar
Quote:
Originally Posted by chaocrator View Post
VA rocks, it packs a lot of features in a small/compact box.
i can sacrifice that smell of analogue to be able to bring my whole rig in a backpack to the gig.
Me too, in all seriousness. I have more VA’s than anything else, I think.
Old 1 week ago
  #7
To OP
You seem to know about dsp's , CPUs and VA s
I would like to know is it that dificult or expensive for one digital synth to have 8 voices instead of 4 ?
Old 1 week ago
  #8
Lives for gear
 
gentleclockdivid's Avatar
 

The only reason why there is confusion about this is because Roland started to use the term for anything that used a subtractive synthesis workflow ( from the gaia onwards ) , the osc's didn't need to be calculated
In the nineties it was simple ,VA stood for digital counterparts of analogue building blocks /subtractive synthesis , pure number crunching , be it novation , yamaha , clavia , roland , waldorf, korg ..all proud and digital and not a sample in sight
The next iteration was in software form , korg with their cmt for the ms20 plugin etc.. , arturia's TAE
Fast forward to the present , albeit it with far more accurate circuit emulation ( u-he , cytomic etc ) , roland ( aira line ) etc..
Blame Roland
Old 1 week ago
  #9
Gear Addict
 

The definition of "VA" isn't the issue, the issue is Roland obfuscating technology with unnecessary marketing buzzwords. WTF is "Zen-Core"? It means absolutely nothing, no more than ACB, because Roland didn't disclose technical details as to how any of these "technologies" work exactly.

If Roland uses FPGA to model circuits then why not just disclose it instead of beating around the bush?

If Roland just generates its oscillators with DSP, then just admit it. If it's just waveforms as PCM on a ROM, then just say it. No need for all these useless marketing gimmicks, in order to fake the fact that there is "something new", there is nothing new with the latest Jupiter from a engineering perspective. When asked about the Peak, Novation has no issue disclosing they use FPGA , for this, DSP for that, and neither do other brands. Only Roland does that. And given the fact that numerous people believe that the boutiques are all analogue, Roland have been extremely successful when it comes to misleading buyers.

Ultimately yes, it's the result that counts, all modelling techniques have trade-off, aliasing, polyphony, sound quality, "faithfulness" of the emulation.
Old 1 week ago
  #10
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
Let me know if you have any questions, or see any egregious errors.
A lot of what you write is nonsense.

Quote:
Analog Modeling: this is just fancy talk for a very complex, CPU-expensive algorithm that can take an input and does give an output. It looks at each component of a circuit for its behavior and then describes it mathematically. By combining all these equations, you get a full model.
You're describing component modeling here, not analog modeling.
Analog modeling can be anything really. The best you can describe it is that is tries, in some undefined way, to model some analog thing. 90's VA's had 'analog modeling' but they for sure didn't model components of a circuit.

Quote:
Oscillator: in the digital world, it's a stream of samples(3) from a sample(2) that describe digitally the shape of the wave normally created in circuitry.
Not all digital oscillators (not even the ones called VA) use stored samples as their base waveform. Not all VA oscillators use waveforms that are normally created in circuitry. There is no strict definition like you make.

Quote:
Band-limited: an oscillator is band-limited when all frequencies higher than ~20Kh are removed. This can be done by building the wave up from so many generated sine waves, or by building bleps through an algorithm that removes unwanted frequencies from a pure waveform. There are other ways, but I just give these as common examples.
-Band limited just tells us that at at some point there are no more higher frequencies. How this is done and at what frequency the sound stops is depending on circumstances. A 96kHz syth could be band limited to 40kHz for instance.
-You're saying that to band limit a signal you have to remove frequencies, but you then give an example of building up a waveform. So, you don't really have to remove (filter) a signal to get a band limited result. You can build it up from basics, like in your example.
-Blep doesn't 'remove unwanted frequencies'. In blep there are no unwanted frequencies in the first place. The point of blep is that you don't generate those frequencies.

Quote:
Interpolation: this is used when transposing samples for playback. There are a number of methods, from the worst--linear to one of the best--sinc. Linear just takes two samples and finds a mid-point. Sinc takes several samples to find a curve to match and then finds the needed point on that curve. Sinc sounds extremely good, but can be very CPU expensive.
-Interpolation is used in many places, not just when transposing samples.
-Linear is not the 'worst' interpolation. One of the most basic interpolations is 'zero hold', which simply holds the value of the previous sample until a new value arrives.
-Sinc interpolation convolves each sample with a particular sin(x)/x function. It doesn't 'find a curve' or any of that other nonsense, that is just the automatic result of the convolution.

Quote:
Back to oscillators: there are two ways of generating them.

One, read a sample(3) from a sample(2) and interpolate as needed if you don't have a sample(2) at the correct frequency. The CPU cost is purely dependent upon the interpolation method used. For single-cycle waveforms, it's very cheap, in both CPU and memory, to just pre-calculate each frequency needed and put them into a lookup table. Then, no interpolation is necessary.

Two: calculate the sample(3) on-the-fly. Depending on the algorithm used, this can be quite expensive in terms of CPU. Pre-calculated lookup tables for sin, cos, tan, atan, etc. can alleviate a lot of that pain.
About method 1, i don't think there are actually any synths out there that 'pre calculate each frequency needed'. Some sample sets for samplers have each note (or every other note) sampled separately but this is not a common method for synths and single cycle oscillators.
One of the problems is that you'd need some special stuff to do things like pitch modulation. To do it properly you would need hundreds of pre-calculated samples between each note.
About method 2, how are the pre-calculated sine tables any different from method 1?
Besides this, there are other methods of synthesizing signals. A lot of methods can't be easily classified as purely 1 or purely 2.

Quote:
Back to filters: filters must necessarily be calculated on-the-fly since they do not know what their input will ever be. So, there are dozens of filter models out there. Some are purely mathematical in their creation and are very CPU-cheap to use. Modeled filters tend to be very CPU-expensive. Diva is the spendy one. Arturia filters are the cheaper ones, probably by simplifying the algorithms for speed; however, the edge cases (extreme values) may not sound as good as U-he's models.
-Filters can also be put into tables. You can even go back one step and make pre-filtered wavetables and not have a filter at all. A lot depends on the kind of filter you want.
-In a computer all filters are purely mathematical, not just some.
-All these filters are models. How much CPU is used depends on how complex the model is. To make a model that behaves more like a particular analog filter you will need to spend more CPU to get the details down.

Quote:
Finally, there are anti-aliasing filters that can be applied, all of varying efficacy vs. CPU-expense. These are usually applied at the very end before the data goes out to the effects and can be used in addition to or in lieu of band-limited oscillators.
No, AA filters are not applied at the end. At the end all aliasing has already occurred and you can't filter it out anymore.
AA filters are typically applied before the operation that would introduce the aliasing.

All in all i think you actually have added to the confusion instead of cleared things up.
But cudo's for effort.
Old 1 week ago
  #11
Gear Guru
 
Yoozer's Avatar
Quote:
Originally Posted by Hysplex View Post
To OP
You seem to know about dsp's , CPUs and VA s
I would like to know is it that dificult or expensive for one digital synth to have 8 voices instead of 4 ?
Twice the processing needed, so twice the processing power needed. Expense is more in things like packaging, though you can have pricy DSPs.

A MicroKorg has 4 voices becuase the design has never beem refreshed. Parts are cheap, R&D is recouped.

Boutiques are deliberately limited. Important difference
Old 1 week ago
  #12
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by BobTheDog View Post
Good post.

Not really an error but it might be a good idea to edit the Band-Limited section to say the frequency may well change depending on the Nyquist Frequency rather than ~20Khz.
I'm speaking about audible frequencies from a sampling rate-agnostic point of view. The most common AA filter is a half-band, which strips out frequencies greater than the Nyquist limit. However, given 96Khz as your sampling rate, if you only strip out above 48Khz, yes, it is anti-aliased. However, if someone comes along and naively downsamples to 48Khz (or worse, 44.1Khz), there's still all those frequencies from 22.050Khz and above that will promptly march down and alias all over everything. Thus, depending upon your use case, it's best to strip from ~20Khz and above. Part of the jiggery-pokery of the magic 44.1Khz sampling rate was to leave a small band from 20Khz to 22.050Khz for the curve of the AA filter since there's no brickwall, per se.
Old 1 week ago
  #13
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by monomer View Post
A lot of what you write is nonsense.
Uh....

Quote:
Originally Posted by monomer View Post
You're describing component modeling here, not analog modeling.
Analog modeling can be anything really. The best you can describe it is that is tries, in some undefined way, to model some analog thing. 90's VA's had 'analog modeling' but they for sure didn't model components of a circuit.
Yes, Analog Modeling can be whatever people want to call it. Which adds to the confusion. True modeling is as I describe. Applying some random cubic or quadratic equation to a sample(3) because it kinda sounds analog is not. It's faking it.


Quote:
Originally Posted by monomer View Post
Not all digital oscillators (not even the ones called VA) use stored samples as their base waveform. Not all VA oscillators use waveforms that are normally created in circuitry. There is no strict definition like you make.
Correct. That's why I defined two only possible ways of generating them: using pre-calculated samples or generating them on-the-fly.


Quote:
Originally Posted by monomer View Post
-Band limited just tells us that at at some point there are no more higher frequencies. How this is done and at what frequency the sound stops is depending on circumstances. A 96kHz syth could be band limited to 40kHz for instance.
It depends on use case. See my previous reply.

Quote:
Originally Posted by monomer View Post
-You're saying that to band limit a signal you have to remove frequencies, but you then give an example of building up a waveform. So, you don't really have to remove (filter) a signal to get a band limited result. You can build it up from basics, like in your example.
-Blep doesn't 'remove unwanted frequencies'. In blep there are no unwanted frequencies in the first place. The point of blep is that you don't generate those frequencies.
Now you're picking semantic nits.

Quote:
Originally Posted by monomer View Post
-Linear is not the 'worst' interpolation. One of the most basic interpolations is 'zero hold', which simply holds the value of the previous sample until a new value arrives.
Point to you.

Quote:
Originally Posted by monomer View Post
-Sinc interpolation convolves each sample with a particular sin(x)/x function. It doesn't 'find a curve' or any of that other nonsense, that is just the automatic result of the convolution.
Which is fancy talk for finding a curve. Think about what a sample(2) models.

Quote:
Originally Posted by monomer View Post
About method 1, i don't think there are actually any synths out there that 'pre calculate each frequency needed'. Some sample sets for samplers have each note (or every other note) sampled separately but this is not a common method for synths and single cycle oscillators.
I'm sort of throwing in VSTi techniques here. I don't know how every VA engine living in a keyboard is implemented.

Quote:
Originally Posted by monomer View Post
One of the problems is that you'd need some special stuff to do things like pitch modulation. To do it properly you would need hundreds of pre-calculated samples between each note.
Not really. Given a reference pitch, minor pitch variation through interpolation is still quite easy and you are less likely to get weird formant-y things than if you worked with samples in fifths (or octaves ) vs just steps.

Quote:
Originally Posted by monomer View Post
About method 2, how are the pre-calculated sine tables any different from method 1?
Calculating a sine is a very expensive mathematical operation compared to looking it up from a table, which can hold better precision than is actually needed. It's a great way to throw memory at a CPU use problem when you are performing live calculations. The DX line used a sin LUT to make the magic happen.

Quote:
Originally Posted by monomer View Post
Besides this, there are other methods of synthesizing signals. A lot of methods can't be easily classified as purely 1 or purely 2.
There really isn't any other way here... Either your samples(3) are generated by reading them from a table or by calculating them. I'm not including any interpolation, etc. here.


Quote:
Originally Posted by monomer View Post
-Filters can also be put into tables. You can even go back one step and make pre-filtered wavetables and not have a filter at all. A lot depends on the kind of filter you want.
Technically, yes. I believe Roland did this in one of their boards, but it's hella inflexible.

Quote:
Originally Posted by monomer View Post
-In a computer all filters are purely mathematical, not just some.
-All these filters are models. How much CPU is used depends on how complex the model is. To make a model that behaves more like a particular analog filter you will need to spend more CPU to get the details down.
I believe that's what I said, but thank you for clarifying.

Quote:
Originally Posted by monomer View Post
No, AA filters are not applied at the end. At the end all aliasing has already occurred and you can't filter it out anymore.
AA filters are typically applied before the operation that would introduce the aliasing.
I left out the step of oversampling. My brain tends to go from A->Z by way of Q, B, X, and M. I'll go fix it...

Quote:
Originally Posted by monomer View Post
All in all i think you actually have added to the confusion instead of cleared things up.
But cudo's for effort.
Trust me. I'm a doctor.

I'm just trying to clear up some basic misconceptions here in general terms.

I hope for no more of this:

ME: (explains technical concept trying to clear up some basic misconception)
SOME RANDO DUDE: It's modeled.
OP: Oh, okay. Thanks, SOME RANDO DUDE, for reinforcing my incorrect belief!

Old 1 week ago
  #14
Lives for gear
 
VennD68's Avatar
Quote:
Originally Posted by camus2 View Post
The definition of "VA" isn't the issue, the issue is Roland obfuscating technology with unnecessary marketing buzzwords. WTF is "Zen-Core"? It means absolutely nothing, no more than ACB, because Roland didn't disclose technical details as to how any of these "technologies" work exactly.

If Roland uses FPGA to model circuits then why not just disclose it instead of beating around the bush?

If Roland just generates its oscillators with DSP, then just admit it. If it's just waveforms as PCM on a ROM, then just say it. No need for all these useless marketing gimmicks, in order to fake the fact that there is "something new", there is nothing new with the latest Jupiter from a engineering perspective. When asked about the Peak, Novation has no issue disclosing they use FPGA , for this, DSP for that, and neither do other brands. Only Roland does that. And given the fact that numerous people believe that the boutiques are all analogue, Roland have been extremely successful when it comes to misleading buyers.

Ultimately yes, it's the result that counts, all modelling techniques have trade-off, aliasing, polyphony, sound quality, "faithfulness" of the emulation.

The RADIAS used MMT (Multiple Modelling Technology).
Yamaha used Proprietary Analog Modelling using DSP VLSI processors and then introduced FDSP to the AN engine in the EX-5
Kurzweil uses V.A.S.T
And so on and so forth.
Virtual Analog was term used in large number of magazines and was simply broader marketing speak for pretty much any DSP based Digital Subtractive Synthesis that by in large didn't use sample based wave forms.
Novation may use FGPA's but they also use NCO's are we going to accuse them of using marketing speaks that leads to misunderstanding what an oscillator is.
At the end of the day one shouldn't care about the approach to synthesis involved.
If an instrument sounds good and has the feature set you want, like and/or gel with and enhances your approach to synthesis and music making, that's all that really matters.
Pretty much any manufacturing sector with end product markets itself how it sees fit and how much of it is really transparent across a broad range of consumer based market sectors.
Its not like obfuscating the issue is anything unique to Roland because it isn't.
We could list an endless number of companies that do the same but why bother ?
Does Gearslutz always have to be a never ending long winded version of Twitter for the synth world ?
Old 1 week ago
  #15
Lives for gear
 
DrJustice's Avatar
 

Quote:
Originally Posted by camus2 View Post
The definition of "VA" isn't the issue, the issue is Roland obfuscating technology with unnecessary marketing buzzwords. WTF is "Zen-Core"? It means absolutely nothing, no more than ACB, because Roland didn't disclose technical details as to how any of these "technologies" work exactly.

If Roland uses FPGA to model circuits then why not just disclose it instead of beating around the bush?

If Roland just generates its oscillators with DSP, then just admit it. If it's just waveforms as PCM on a ROM, then just say it. No need for all these useless marketing gimmicks, in order to fake the fact that there is "something new", there is nothing new with the latest Jupiter from a engineering perspective. When asked about the Peak, Novation has no issue disclosing they use FPGA , for this, DSP for that, and neither do other brands. Only Roland does that. And given the fact that numerous people believe that the boutiques are all analogue, Roland have been extremely successful when it comes to misleading buyers.

Ultimately yes, it's the result that counts, all modelling techniques have trade-off, aliasing, polyphony, sound quality, "faithfulness" of the emulation.
This is my daily posting on this subject (it's starting to feel like repetition and belabouring by now - it annoys me and probably others too...)

Roland has not claimed to be using FPGAs, instead they've told us that they use their own DSPs (ESC/BMC). Not that it matters a jot, but they're clear on this.

They've also let us know that in the Zen-Core engine, the oscillator type is selectable as one of PCM, VA, PCM-Sync, SuperSAW and Noise. Read about it in the new Fantom manual(s), e.g. here.

As for ACB, that means Analogue Circuit Behaviour, so modelling on the component or circuit-block level, using a DSP. We have known this for years, you can read about it e.g. here. No company will disclose "exactly" how they implement their tech - do not expect them to publish the source code or the schematics of their chips.

Roland is a big (in synth terms) Japanese company, of course they use marketing terms. So do many others. "ASM", "Oxford Oscillators", "VAST", "TAE", "VS", "Z-Plane"... Do a bit more reading than the headlines and you'll find the info. I really don't see the problem.

I find it highly unlikely that Roland is running a nefarious scheme to subvert the world of synths by being secretive about these things (that aren't secrets at all...).

For those that believe that the ACB Boutiques are analogue... well... There are people who believe the DX-7 is analogue too. That's not Roland's or Yamaha's fault.

BTW, this may be the wrong thread for pure Roland bashing - head over to the X/Xm thread for the state of the art in that area
Old 1 week ago
  #16
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
Trust me. I'm a doctor.

I'm just trying to clear up some basic misconceptions here in general terms.

I hope for no more of this:

ME: (explains technical concept trying to clear up some basic misconception)
SOME RANDO DUDE: It's modeled.
OP: Oh, okay. Thanks, SOME RANDO DUDE, for reinforcing my incorrect belief!

Hah, thanks for taking it so lightly.

Well, to address your concerns about people not understanding things, maybe it would be more helpful to make a proper dictionary kind of thing.
What you wrote can be seen as helpful from some points of view but IMO it is limited in scope (not general enough). I think people having questions rarely would see things the way you explained them.
On the other hand, too technical information is also not useful for people unfamiliar with the territory.

This is maybe not so much a critique of your writing but more about finding a way of explaining things in a way that is more coherent and interrelated. In that respect i understand why you wanted to write these things.

So maybe we need a slow and more detailed exposition of some of the basic ideas and maybe first explain what the problems are and how a particular thing (technique, algorithm, philosophy) tries to fix it. For instance, learning what anti aliasing is is almost pointless if you don't already know what aliasing is. A lot of this stuff is also not specific to VA so there is a lot of space for generalization.

There was an effort some times ago here on GS to put together some sort of dictionary thing. I'm wondering what happened to it.??
Old 1 week ago
  #17
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by monomer View Post
Well, to address your concerns about people not understanding things, maybe it would be more helpful to make a proper dictionary kind of thing.
What you wrote can be seen as helpful from some points of view but IMO it is limited in scope (not general enough). I think people having questions rarely would see things the way you explained them.
On the other hand, too technical information is also not useful for people unfamiliar with the territory.
Unfortunately, not everybody thinks the way I do (God help us all if they did!) or the way that you think, either. It's why I chose a thread rather than plunking it down on a monolithic web page. Admittedly, I did thumbnail sketches of everything, but I'm hoping to just inject some concepts into people's brains and let them grapple with understanding them on their own. I'm happy to edit the OP to clarify bits as questions come up, but I'd rather refer people on to resources that can fully explain a single topic. DSP is an YUGGGGEEEE rabbit hole to go down. I'm still wrestling with the math of filters. It's why I hated theoretical physics so much--if I can't grok the real-world applications of the math, I can't really grok the math.

I could also do a post on what logical fallacies to avoid. Trust me, I'm a doctor!
Old 1 week ago
  #18
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
I'm speaking about audible frequencies from a sampling rate-agnostic point of view. The most common AA filter is a half-band, which strips out frequencies greater than the Nyquist limit. However, given 96Khz as your sampling rate, if you only strip out above 48Khz, yes, it is anti-aliased. However, if someone comes along and naively downsamples to 48Khz (or worse, 44.1Khz), there's still all those frequencies from 22.050Khz and above that will promptly march down and alias all over everything.
What you describe applies to any form of downsampling. That's why you need to filter the source before the downsampling and shouldn't be doing naive downsampling.

The 20kHz is arbitrary from a signals perspective. Your example is equally valid for a 192kHz signal that is downsampled to 96kHz. If you don't want aliasing you can filter at ~40kHz and all will be well. This 20kHz just happens to be a common cutoff rate because a lot of converters are 44.1kHz or 48kHz.
Main thing is that AA has to do with the frequency components in the signal at whatever samplerate and not with the hearing limit of humans.


Quote:
Thus, depending upon your use case, it's best to strip from ~20Khz and above.
This 20kHz stripping is only 'best' if the destination sample rate is close to 40kHz.
What frequency you start cutting (and by how much) is related to sampling theorem and the nyquist frequency, not the limit of human hearing. It is a technical requirement for avoiding aliasing.
Old 1 week ago
  #19
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
but I'm hoping to just inject some concepts into people's brains and let them grapple with understanding them on their own.
Then i think at the very least you should try to put things in a way that won't lead people astray and doesn't introduce more myth into an already murky field.
Which is a big challenge, i agree.. Especially the business of making things as simple as possible but not simpler.
A crooked abstraction can keep you occupied for too much time before it becomes obvious what really is going on. At least, that is my experience. The number of myths i had to punch through on the subject of audio throughout my life...
Old 1 week ago
  #20
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by monomer View Post
That's why you need to filter the source before the downsampling and shouldn't be doing naive downsampling.
Unfortunately, unless I write the downsampler myself, I don't know what it's actually doing. If said piece of software is downsampling from, say, 96khz to 48khz, is it just throwing away every other sample or is it averaging each pair (a simple 1-pole LPF) or is it running it through a half-band polyphase filter or is it doing its own thing based on the ratio of old sampling frequency to new and interpolation AND is it filtering at all?

Considering that audible frequencies over ~20Khz can only be heard by those few humans in isolation, I contend it's better to set your AA filter to cut around there as opposed to anything higher and save some future grief.
Old 1 week ago
  #21
Gear Guru
 
Yoozer's Avatar
Quote:
Originally Posted by xanderbeanz View Post
VA sucks, it never got the smell of Analogue right.

They're not allowed to use those kind of fire retardants anymore. Too bad, it really is something special.

Quote:
Originally Posted by camus2 View Post
The definition of "VA" isn't the issue, the issue is Roland obfuscating technology with unnecessary marketing buzzwords. WTF is "Zen-Core"? It means absolutely nothing, no more than ACB, because Roland didn't disclose technical details as to how any of these "technologies" work exactly.
True.

Quote:
If Roland uses FPGA to model circuits then why not just disclose it instead of beating around the bush?
Because it's ultimately something for corksniffers. Pick the best algorithms and the worst DSP, and you have an Access Virus. Pick the worst algorithms and the best DSP, and you've got a... well, I shouldn't say it, the owners will probably be mad at me for doing so

Though - in case of the SH-32 and Gaia it was misleading, because their use of samples had a very noticeable effect on the supersaw. Enabling oscillator sync on the Gaia made it mono.
Old 1 week ago
  #22
Lives for gear
 

This thread is going in the wrong direction. The core issue is that there's no agreement about what "VA" even is.

As far as synths go, the Nord Lead coined the term "VA". Its goal was to offer the sound and functionality associated with analog synths, but more reliable and cheaper. It did so using modern digital technology (and a knobby interface). This is what other manufacturers mean and do to, although the exact digital methods vary. They're describing what something is rather than how they're doing it.

Problem is that the key criteria many use for a synth to count as a "VA" is that its digital (duh) and its oscillators are generated on the fly, "math", rather than sampled, "data". They don't care how the rest works. Here the "how" is what defines the category.

Anyway, don't want to spend too much time on this. I'll drop this again, it just sums up the situation nicely:

M-Audio's Venom was a neat low cost VA that came and went. It had 12 voices and used sample based oscillators, something Taiho Yamada (Venom's lead, long time sound, and later synth, designer) wasn't too shy about. Here's what he had to say about Venom's VA classification (from Matrixsynth's Venom review):

1. What is Virtual Analog Modeling in Venom?
"Ah, Analog Modeling… These days it really could mean that you’ve used DSP to model everything down to the component level, but back when the Nord Lead first came out, it seemed like it meant, “We used DSP to make something that approximates the sound of an analog synth.” In 1995, I doubt they had the processor power to make anything close to a perfect model, and sure enough the Nord Lead really didn’t sound like any analog I’d heard before. But it still sounded awesome, and started the whole Virtual Analog category that came afterwards. Working with, and speaking to, engineers over the years, I’ve heard some of the tricks they use to generate synthesizer voice paths and you’d be surprised how many corners they cut in order to maximize efficiency in imbedded systems. There are modeling synths out there from various manufacturers that probably wouldn’t be considered modeling if people knew exactly how the sound was generated. Keeping that in mind, I think that having a perfectly, technically accurate, model is really not the ultimate goal. All that matters is how close the synth sounds to your target. It’s all 1s and 0s in the end anyway. Does it really matter how those 1s and 0s were generated? Well, I guess it kinda does. Venom’s sample based oscillators buy us excellent timbral accuracy, but almost no behavioral accuracy. We have to re-introduce analog movement after the fact. In a fully accurate turnkey model, you would build all that into the oscillator and have it there automatically. However, I wanted to have the option for digital behaviors in Venom as well, so we broke the algorithms out into the Start Mod and Drift parameters where the level of analog behavior could be controlled. For extra emulation of analog drift over time, I feel it is necessary to add a Smooth Sample and Hold modulation with different amounts routed to the individual oscillators. And for a super geeky level of accuracy, I would then hit the first Sample and Hold Rate with another Smooth Sample and Hold modulation so that the timing of the drift is randomized as well. In regard to the Filter, we put in a tube saturation limiting algorithm to control digital clipping while maintaining the maximum system level and providing an additional place to overdrive the signal. Venom doesn’t emulate any particular analog filter, but tries to capture some useful functionality from across the analog world. That said, if I pair the 12 dB filter with the OB waves, or the 24 dB LP with the MG waves, the synth really does sound like my SEM and Minimoog respectively. I can definitely still hear the difference, but it’s pretty close, and yet still somehow maintains a sonic quality unique to Venom. BTW, no one ever mentions the VCA, but that’s also important to the overall sound of a synth. Again, we don’t try and recreate any particular analog circuit, but I am conscious of how the VCA works along with the envelopes to make a synth sound punchy like a Mini, or chewy like an MS-20. The emulative sounds in Venom definitely take things like minimum envelope rise times into consideration. To sum up, I would say that I feel perfectly comfortable calling Venom a “Virtual Analog Synthesizer” as is silkscreened on the top panel. There is a wide degree of variance in both the technical and sonic accuracy of the models out there – remember the Nord Lead that started it all. Also keep in mind that M-Audio never says Venom is an “analog modeling” synth, although I feel that definition wouldn’t be entirely inaccurate. I guess it comes down to what the most accepted definition of an analog modeling synth actually is…"
"Virtual Analog" is not "Analog Modelling" although maybe it is (but then that's different from "Circuit Modelling"...is "Circuit Modelling" a form of "Physical Modelling"...)

"Does it really matter how those 1s and 0s were generated?"
Yup, it makes a huge difference in how much people are willing to listen to or pay for a synth. Same thing as Digital VS Analog really. Or through-hole vs SMT and so on.
Old 1 week ago
  #23
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by ToyBox View Post
M-Audio's Venom was a neat low cost VA that came and went. It had 12 voices and used sample based oscillators, something Taiho Yamada (Venom's lead, long time sound, and later synth, designer) wasn't too shy about. Here's what he had to say about Venom's VA classification (
I'll take the Venom's take on VA over, say, the Blofeld any day. It's so alive and characterful. It's a pity it was hobbled the way it was.

Unfortunately, VA is an overloaded term, but for the sake of this thread, let's define it as a digital synth trying to sound analog. You have your pure VAs, like the Nord A1, and then there's everything up to the new Jupiter-X, which is essentially a glorified rompler with "VA" components.

In the end, regardless of what jiggery-pokery goes on under the covers, the important question is how does it sound?
Old 1 week ago
  #24
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
I contend it's better to set your AA filter to cut around there as opposed to anything higher and save some future grief.
Yes, i understand this, but it is an opinion.

There is nothing wrong with an opinion but it is not a general fact about band limiting, anti aliasing filters, etc.
I think you shouldn't murk things up with opinion if you haven't explained basic like what is aliasing or how one interpret samples at different rates yet. There is just not enough context and so saying things like that creates what is basically a myth.

If this was the first explanation i had read i'd think that anti-aliasing is just a 20kHz filter applied for mystery reasons.

I think in general there is just a lot of context missing. Maybe it's just not a good idea to explain such terms in isolation...
Old 1 week ago
  #25
Lives for gear
 
syntonica's Avatar
Quote:
Originally Posted by monomer View Post
Yes, i understand this, but it is an opinion.

There is nothing wrong with an opinion but it is not a general fact about band limiting, anti aliasing filters, etc.
I think you shouldn't murk things up with opinion if you haven't explained basic like what is aliasing or how one interpret samples at different rates yet. There is just not enough context and so saying things like that creates what is basically a myth.

If this was the first explanation i had read i'd think that anti-aliasing is just a 20kHz filter applied for mystery reasons.

I think in general there is just a lot of context missing. Maybe it's just not a good idea to explain such terms in isolation...
I just changed it to "frequencies above audible range" and will leave it at that. I know it sort of begs the question, but I don't want to try to explain when/why/how frequencies fold back as its beyond the scope of this thread.
Old 1 week ago
  #26
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by syntonica View Post
but I don't want to try to explain when/why/how frequencies fold back as its beyond the scope of this thread.
But then what is the point of talking about band limiting or AA filters?
Old 1 week ago
  #27
Lives for gear
 
xanderbeanz's Avatar
Quote:
Originally Posted by Yoozer View Post
They're not allowed to use those kind of fire retardants anymore. Too bad, it really is something special.


True.


Because it's ultimately something for corksniffers. Pick the best algorithms and the worst DSP, and you have an Access Virus. Pick the worst algorithms and the best DSP, and you've got a... well, I shouldn't say it, the owners will probably be mad at me for doing so

Though - in case of the SH-32 and Gaia it was misleading, because their use of samples had a very noticeable effect on the supersaw. Enabling oscillator sync on the Gaia made it mono.
My Mono/Poly smells like charged particles and sawdust.
Old 1 week ago
  #28
Lives for gear
 
Arglebargle's Avatar
I love the smell of charged particles in the morning....
Old 1 week ago
  #29
Lives for gear
 
robinkle's Avatar
The term VA (virtual analog) only makes sense when simulating analog behavior. Otherwise it’s just a digital oscillator.
Old 1 week ago
  #30
Lives for gear
 
monomer's Avatar
 

Quote:
Originally Posted by robinkle View Post
The term VA (virtual analog) only makes sense when simulating analog behavior. Otherwise it’s just a digital oscillator.
Define 'simulating'.
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump