Cherry Audio Forums

Hello @MRBarton,

Just bought the VM900 collection. What incredible sounds. Thank you for your hard work and dedication.

Now to spend the next year learning it!

andro wrote: ↑Mon Dec 06, 2021 3:38 pm Just bought the VM900 collection. What incredible sounds. Thank you for your hard work and dedication.

After my 7-day demo ran out, it was a no-brainer for me. I reconstructed the Moog I used in college (early 1970s) and the VM900 brought back very happy memories in terms of the sounds and the workflow. But it is so much easier to save and restore patches. No need to put up a sign "Please don't touch!" when I had to take a break for classes. If I recall correctly, the model we had at our disposal cost around $3000. But the >$150,000 figure seems accurate in terms of contemporary dollars and the additional modules which we didn't have. I have no idea how much comparable modules would cost today.

I am using the 'scope from the VM2500 which very much reminds me of the one we had in the Moog studio. Any chance to get the VM2500 'scope reskinned with the VM900's design?

andro wrote: ↑Mon Dec 06, 2021 3:38 pm Now to spend the next year learning it!

Not a bad way to spend one's time!!

MRBarton wrote: ↑Fri Dec 03, 2021 1:22 am VCV Rack and Voltage Modular have no "sound". What you are saying is that one empty cabinet sounds better than the other, and that's all these systems are; empty cabinets. VCV and VM are infrastructure that pass on what the modules sound like. You have to decide on sound quality, desirability, etc. module-by-module. The sound comes from the modules, not the system. So it is up to you to decide which developers you like and will trust in the future. I'm sure quality varies all over the place everywhere, but it is not the fault of either system, therefore it makes no sense to say that either is better. You might like one over the other for other reasons, but sound should not be one of them.

While this is certainly true to a large extent, the more I use VM the more I think it would benefit from some kind of global oversampling to reduce aliasing, which sort of falls under the banner of Voltage Modular having a "sound" (a global way of processing audio). When dealing with something as harmonically rich as sawtooth or square waveforms, proper AA really is not optional. Same goes for processes that generate harmonics like FM, AM, distortion & other nonlinear processes. But AA still seems to be missing entirely from many modules, or lousily implemented in others -- including some of Cherry Audio's own modules.

We'd have to rely on trust a lot less with some global 4x oversampling. CA's recent updates to the standalone synths to add oversampling options gives me hope that something of the sort will be added to VM in the future. If it is, it will allow me to use a lot of modules I've been avoiding, and to buy some new ones which are currently unusable to me because of unacceptable levels of aliasing.

(The VM900 VCOs are practically alias free -- well done!)

jkeller51 wrote: ↑Tue Dec 07, 2021 2:49 am

MRBarton wrote: ↑Fri Dec 03, 2021 1:22 am VCV Rack and Voltage Modular have no "sound". What you are saying is that one empty cabinet sounds better than the other, and that's all these systems are; empty cabinets. VCV and VM are infrastructure that pass on what the modules sound like. You have to decide on sound quality, desirability, etc. module-by-module. The sound comes from the modules, not the system. So it is up to you to decide which developers you like and will trust in the future. I'm sure quality varies all over the place everywhere, but it is not the fault of either system, therefore it makes no sense to say that either is better. You might like one over the other for other reasons, but sound should not be one of them.

While this is certainly true to a large extent, the more I use VM the more I think it would benefit from some kind of global oversampling to reduce aliasing, which sort of falls under the banner of Voltage Modular having a "sound" (a global way of processing audio). When dealing with something as harmonically rich as sawtooth or square waveforms, proper AA really is not optional. Same goes for processes that generate harmonics like FM, AM, distortion & other nonlinear processes. But AA still seems to be missing entirely from many modules, or lousily implemented in others -- including some of Cherry Audio's own modules.

We'd have to rely on trust a lot less with some global 4x oversampling. CA's recent updates to the standalone synths to add oversampling options gives me hope that something of the sort will be added to VM in the future. If it is, it will allow me to use a lot of modules I've been avoiding, and to buy some new ones which are currently unusable to me because of unacceptable levels of aliasing.

(The VM900 VCOs are practically alias free -- well done!)

Hey, whadya mean "practically"?? Just kidding.

You might like to know that the VM900 oscillators are not oversampled at all, so oversampling is not always a panacea. My Laboratory VCOs are oversampled quite a bit and alias more. I will be updating them to my new technology soon (almost done). Every filter in VM2500 and VM900 is oversampled as well as the Oberheim filter.

Global oversampling is not practical to add. What happens inside a module is up to each developer to take care of. If the system ran faster, everybody would have to rewrite everything, and if they didn't, it would sound the same because no system can undo aliasing once it's there. This goes to my original point of VM having no "sound". Each dev is free to oversample (which is not hard to do) or not.

Sometimes oversampling is needed and sometimes it is a waste of CPU. Imagine if a sequencer or LFO had to work 4x harder because the whole system is running at a higher sampling rate -- a complete waste. I believe the way things are is best and the original cooks knew what they were doing. That's why there is a trial period on everything so the pickier among us can make up their own minds product-by-product. It's a given that the level of DSP expertise among the devs varies quite a bit. There are pros and hobbyists in the store and that's great. In the few years I've been doing VM stuff, my own chops have improved dramatically. That's why I'm going back to revisit some of my own modules to make them better. Who knows, maybe one day soon, a so-called "hobbyist" will astound our socks off.

I hope that sheds some more light on this topic. I read these forums every day to answer any questions you may have. If anyone wants to send me a private question or just tell me how much I'm loved, [email protected] is the place.

--mb

fascinating conversation, thank you!

My pleasure.

MRBarton wrote: ↑Tue Dec 07, 2021 4:01 am Hey, whadya mean "practically"?? Just kidding.

You might like to know that the VM900 oscillators are not oversampled at all, so oversampling is not always a panacea. My Laboratory VCOs are oversampled quite a bit and alias more. I will be updating them to my new technology soon (almost done). Every filter in VM2500 and VM900 is oversampled as well as the Oberheim filter.

Global oversampling is not practical to add. What happens inside a module is up to each developer to take care of. If the system ran faster, everybody would have to rewrite everything, and if they didn't, it would sound the same because no system can undo aliasing once it's there. This goes to my original point of VM having no "sound". Each dev is free to oversample (which is not hard to do) or not.

Sometimes oversampling is needed and sometimes it is a waste of CPU. Imagine if a sequencer or LFO had to work 4x harder because the whole system is running at a higher sampling rate -- a complete waste. I believe the way things are is best and the original cooks knew what they were doing. That's why there is a trial period on everything so the pickier among us can make up their own minds product-by-product. It's a given that the level of DSP expertise among the devs varies quite a bit. There are pros and hobbyists in the store and that's great. In the few years I've been doing VM stuff, my own chops have improved dramatically. That's why I'm going back to revisit some of my own modules to make them better. Who knows, maybe one day soon, a so-called "hobbyist" will astound our socks off.

I hope that sheds some more light on this topic. I read these forums every day to answer any questions you may have. If anyone wants to send me a private question or just tell me how much I'm loved, [email protected] is the place.

--mb

Reviewing my notes, the VM900 VCO has no discernable aliasing whatsoever above or near the "analog" noise floor, just to clear that up! It's impossible for me to tell if there is aliasing, say, 20 dB below that, hence "practically" alias free

I actually suspected the VM900 VCOs are not oversampled -- the way the upper harmonics drop off in batches twice per octave as the pitch rises looks strikingly similar to my own oscillator implementation in software elsewhere, which does not use oversampling but is based on wavetables of perfectly bandlimited harmonic series. So I agree totally that oversampling is not always the best solution -- but even a brute force solution is better than none, IMO.

It's true that not all modules require AA, but a solution where half the modules are individually upsampling, processing, then downsampling before passing the signal on to the next module to be upsampled, processed, then downsampled doesn't seem exactly CPU efficient either. It also puts a lot of onus on the developers, who, as you point out, vary in their DSP knowledge and maybe want to focus on the fun stuff and not boring things like interpolation and decimation.

Anyway, I was just suggesting a way to improve the audio quality across the board without expecting devs to figure out how to implement AA properly for their modules, since many are apparently not concerned with it. The most common thing I've seen VM dinged for across the web, especially compared to Softube Modular or Reaktor, is sound quality. I can't help but think the heavy aliasing of many modules as well as fairly common transients (e.g. stemming from non-bandlimited CVs) play a big part in this. I don't have the competitors' products, so I don't know how they handle it.

<<I actually suspected the VM900 VCOs are not oversampled -- the way the upper harmonics drop off in batches twice per octave as the pitch rises>>

Ooh, aren't you the detective!

I'll agree with absolutely everything you said, but we walk a tightrope between performance and CPU efficiency. I personally think everything sounds great as it is with the very occasional roughness around the edges which can usually be worked around or just disappears in a mix. If everything ran at 192KHz, you could only run a quarter as much before your CPU hits the wall and I don't think anyone would be happy with that. Have you seen the videos of some of the absolutely gigantic patches people are making? Ya pick your poison.

I develop on a machine that is a decade old just to make sure that everyone will be able to run everything without a problem.

--mb

An interesting conversation about over-sampling.

I am one of those people wanting to build larger and larger patches. The demo tune I'm working on at the moment uses 95 modules and counting and I consider it quite modest in scale yet I have to be careful in order to keep the CPU load low (currently about 15% on my machine).

As an aside my dream would be to be able to compartmentalize patches and elegantly switch on processing in some parts and off in others. An improved version of the plug-in host could do this and I'm toying with developing this myself but would love CA to mod their own host code instead of me wasting time going down a route that involves a duplication of effort.

Being forced to switch from 48 kHz to 192 kHz is definitely something I don't want to see happen as it would mean the kind of work I'm interested in wouldn't be feasible on average machines.

My current modules are in the CV domain but that doesn't mean that heavy duty computation isn't involved nor that high-frequency response isn't important but 48 kHz seems like a good basic sample rate.

I'm also now doing R&D in the audio domain and interpolation rather than over-sampling seems to be the key thing that matters in the area I'm looking at.

On the broader topic of VCV Rack, as a user I would like to see some features in VCV Rack added to VM. Having the option to interactively check the CPU load of individual modules would be great as would offline rendering.

As a developer my main problem with VCV Rack it is that it uses C++ rather than Java. I was an early adopter of C++ and in the late 80's it was a big improvement but that's thirty odd years ago and C++ now looks decrepit.

Finally, I really would like to see CA focus more on VM rather than VSTs. There are thousands of VSTs to choose from but VM is something special.

A switch would make everybody happy... except the developers

ColinP wrote:As a developer my main problem with VCV Rack it is that it uses C++ rather than Java.

Ohhh now you're selling me on VCV rack.