DSP Robotics Support

The Quilcom SECTION is a synth based on my experiments to create a more authentic instrument ensemble sound.
Classic methods of simulating this include chorus effects which mix a live signal with at least one delay with slow time modulation, and/or to mix several oscillators playing in detuned unison and panned left to right. Both these methods produce interesting and useful sounds, but suffer from a very artificial flanging or phasing effect due to being based on using the exact same source waveform which can cause beating as they drift in and out of phase.

The SECTION uses several methods to achieve a multiplicity, the most interesting being what I call Phase Scrambling or randomisation, which I’ve found no mention of anywhere in my searches. If anyone here can find a mention of the technique I’ll cease to assume that it’s my invention.

The system takes a waveform and extracts its partials’ magnitudes via FFT. A range of random phase values is generated then passed into iFFT along with the derived magnitudes. This produces (9) waveforms that have broadly the same timbre but completely different waveforms. Having different waveforms means a very much reduced beating and interference between them, especially when detuned. The sound also has a nice spacial quality to it since the phase scrambling may be similar to effects caused by real environmental factors.

This doesn’t achieve the Holy Grail of totally authentic multiplicity; you’d need real samples of instrument sections for that. I do think it makes for a nice and different sound though, and lends itself to strings and brass very nicely. You could think of the SECTION as a glorified brass and string synth, which can do other stuff too.

I go into more detail in the User Guide included in the zip. You also get the fsm, 32 bit and 64 bit plugins and loads of free waveforms from the Adventure Kid.
Here’s my YouTube video:
https://youtu.be/PcOc-byyLsY

Get it from Dropbox:
https://www.dropbox.com/s/8ug8omg9kv4v9 ... 0.zip?dl=0

I hope you find it interesting and maybe useful, and I would love a few comments please.

Cheers

Spogg

Beautiful. Thank you! Gonna test it deeply later.

Hello Spogg,

Super, another really great synth, I like all the sounds. You did a fantastic job ! Still do not completely understand fft and ifft, especially here, but I will keep studying the schematic and read your guide again.

Later then, BobF.....

After about an hour with Section, I'd like to give some feedback.

The number one issue I have with it is the overall sound. Each and every preset sounds dull/muffled, there's no air. As if the synth engine removes all high frequencies. Many times I went straight to the low pass filter to turn it down, just to realize that there is no filter active.

The least issue I have with Section is this: The pizzicato preset is not pizzicato. It is not even spiccato. It is just a slow attack bowed string cut off quickly.

In between those two are the following issues.

The UI (note the missing G, it's not about graphics) is counter intuitive. I can't find what I'm looking for, unless I spend 20 seconds looking closely at each of the knobs and sections (no pun!). And that each time, I'm looking for something else! There is no relationship pointed out, there's no meaningful seperation of the various sections, there's no user guidance. I really tried to create my own preset, but it was so cumbersome, that I gave up after 10 minutes or so. Sometimes, less is more. Everything on one page is too much information at once for a more complex synth like this one!

The cpu load is surprisingly high. I tested against a synth with 3 totally seperate oscillators (read: 3 synths in one, each has its own filter, lfos, effects, etc). All oscillators set to 32(!) voice unison, and the global voice managment set to 32 voices. Playing 5 notes with such a very loaded preset uses as much cpu as 5 notes played with your church organ preset.

The response of the user interface is sluggish. I don't exactly know why, because there seem to be no animations running. And the knobs seem to move smoothly. Yet, when using the ui to create a preset, it feels as stated above.

I noticed a very high (and unusual) latency between pressing a key and the actual sound producing on some presets (don't ask me, which ones). If it were present on all presets, I would have thought it's the price of using realtime fft, but since it just occurs on some presets, it may be a technical issue?

For really ensemble-ish sounds, the synth is lacking density. The piano preset in lower registers, way below C3, was the only preset, where I really saw a use case of this technique. It sounds so much more realistic (in that lower range) than many multisampled, several gigabyte Pianos. If it just weren't so muffled!

Overall I'm not impressed. That of course doesn't mean, your work isn't appreciated. But the results don't excite me as much, especially considering the complexity that went into this.

Thanks for the comments so far guys!

@tulamide
I especially appreciate you taking the time to give it a proper test drive and additionally for providing such a detailed report. Your comments are always instructive, constructive and very welcome. I’ll respond in the order you made them:

Dull/muffled sound: I did notice this and it seems to be an effect of the FFT to iFFT conversion process. Internally the waveform table is 256 steps before being transformed. I tried higher values but the system took ages to create the new waveforms and was unusable for practical purposes. So I think the higher partial values were adversely affected. You can hear this best on the SUPER SAW preset when you switch between RANDOM PHASE and ORIGINAL WAVE (no conversion). If you choose FIXED PHASE you can still hear the slight loss of high frequencies. This suggests the issue is with the two-way conversion rather than the phase scrambling itself. Ultimately I’m limited to the Prims available to me at the moment.

Pizzicato: you’ve exposed my paucity of knowledge of musical terms!

Confusing UI not easy to navigate: I confess that I made a bad decision. I realised I could fit it all onto one screen, so I did just that. You’ve explained to me before about this so I have no excuse this time. I would say in partial (!) defence that, like with any new synth, you’d get to know it quickly, but that’s not the point. If it’s unwelcoming it’s off-putting.

CPU very high: Guilty as charged. The organ preset uses about 7% above the baseline for every 4 notes pressed. The baseline is affected by the MVERB 7B and cross delays and is around 4% in total. So each wave file reader is using about 0.75% (this is Martin’s improved one) but there are 9 running. Once again, I’m limited by what I have available to use. I imagine your comparison was with a synth coded in C++ or similar, and I could never compete with it. If it was a synth made in FS I would love to see it! If anyone here can see opportunities for further optimisation it would be most appreciated.

UI sluggish: Any change that requires all waves to be updated will cause this behaviour. This could be loading a new base wave or operating the NOISE knob. Changing a preset will also create a slower response. I started the development with 1024 step wave files and it was unusable. 512 was still slow so I had to settle on 256. Using the Wavetable prims was also painful. I didn’t notice a sluggish behaviour on any other settings however, but I have a corei7 system so maybe it’s down to that.

Latency varies with presets: I think this is a sound design issue. The synth can randomise note-on and note-off gates and also Attack times (the amount is adjustable). You can’t look ahead to a key press, so it’s only possible to provide delay to onset, not onset earlier than the beat. If a preset has too much latency you can reduce the random variations for Attack and Gate on.

Lack of density: I would have liked more oscillators but the CPU would be far too high. When I was developing it I tried more, and the sound was denser. So 9 was a reasonable maximum for this synth. I feel that it works better for some instrument sounds than others. Also the illusion of multiplicity varies. For example I think the trumpet and sax sounds are more convincing as sections than that piano preset which sounds like one piano rather than several (it was tuned for the lower register).

Now a confession: I was amazed but disheartened by Martin’s wonderful Chorous effect. It achieves a very similar outcome but for any audio input. I decided to release the Section because it was finished, but I would recommend using the Chorous over the Section.

Cheers

Spogg

I see! It is mostly technical limits. Flowstone might have reached the high end of the scale of what it's capable of, with this synth. Don't be discouraged though, it is still a valuable source of learning for so many Flowstoners here!

Offtopic but probably interesting:
The synth I used to compare the CPU load is this one (video link below). It is a new synth, that during Halloween sale (approx. until 04/11/2019) has a special introductory price reduced by 95%. I bought it for €6!

I normally stay away from the business, but a brand new synth (somewhat limited if you look more closely, but still impressive) as a bargain? I just couldn't resist!

https://youtu.be/ahDxLqVlGGI

That Union synth sounds and looks amazing, and what a bargain! I’m not surprised you invested.
I imagine that at least some of the software is written at machine code level. Or can a C++ compiler achieve that and produce optimised code without intervention?

I guess the approach with FS is to try and work with the inevitable imposed limitations (unless your name happens to be Martin Vicanek!) and see what can be achieved.

With reference to one of my points above, how did you get on with Union as a new and unfamiliar synth? Looking at the UI some of it looked quite puzzling, without referring to a user guide at least. For example, I didn’t spot a master volume control on the video…

Anyway, once again, thank you so much for all that feedback.

Today I’ve been experimenting with Phase Distortion, which is a whole fascinating area to check out. We shall see! If I decide to make a PD synth it will certainly have Martin’s Chorous in it.

Cheers

Spogg

First, union synth require 8gb ram machine minimum! that's insane. Moreover, their users complain that the CPU usage is also very high (I saw the comments on their video) so spog's synth still wins in that category. But yet - comparing between spogg's work (or any other FS synth) and union synth is clearly unfair when we are talking about sound quality and I believe it's not spog's (and our's) fault at all. When I first released my 3 oscs synth back then few years a go, the no. 1 claim against it was a faulty voice management. And I see the same problem repeats in other's releases (dj brynt's synths for instance). The problem is that the sound is powerful, reach with details and nuances - as long as you don't play 3-4 notes chords. Then you take a chord and you end up with noisy siren a like buzz at the background. I have no idea why it happens and it doesn't happens in synthedit or other platforms a like. I remember that it has been discussed on synthmaker's forum many years a go, but somehow, this behavior is still with us. When working with samples, it's less noticeable and that's the good news. But when you go to wavetables, it becomes a disaster unfortunately. A filter can somewhat help to reduce this annoying behavior, but then you loose your "air". And that's not cool. especially when you develop a vsti that aims to simulate strings and other airy instruments.

kortezzzz wrote:First, union synth require 8gb ram machine minimum! that's insane.

Where did you get that from? It's nonsense. It uses a few MBs, just like any other plugin in my library.

kortezzzz wrote:Moreover, their users complain that the CPU usage is also very high (I saw the comments on their video) so spog's synth still wins in that category.

I am a user and can say for sure that it uses just as much CPU on similar settings as, for example, Surge.
Moreover, if you were right, you would play directly into my hands, since my argument was, that I set all 3 oscillators to 32 voice unison (that's a total of 96 per note, just so you may understand it better). Each package of 32 voices were furthermore routed throuh a filter on their own (for a total of 3 filters per note), and 3 effects (for a total of 9 effects per note). This all then got routed to 4 global effects (a compressor, a delay, a chorus and a reverb). Then 5 notes were played simultanously. So at any one time durin those 5-note-play, there was a total of 480 voices, 15 filters and 49 effects active.
And all that used just as much CPU as 5 notes of the church organ preset. You may not like it. But that's the truth. And it was a lot of CPU load on both synths. And I mentioned it, because I use other Quilcom synths, who are not nearly as CPU intensive.

kortezzzz wrote:But yet - comparing between spogg's work (or any other FS synth) and union synth is clearly unfair when we are talking about sound quality and I believe it's not spog's (and our's) fault at all.

I didn't. The only thing I did was comparing CPU loads of both synths.

Spogg wrote:Or can a C++ compiler achieve that and produce optimised code without intervention?

Yes. I don't know it for sure, but I'd bet it was programmed with JUCE, the de-facto standard of plugin creation nowadays. JUCE is a C++ library.

Spogg wrote:With reference to one of my points above, how did you get on with Union as a new and unfamiliar synth? Looking at the UI some of it looked quite puzzling, without referring to a user guide at least. For example, I didn’t spot a master volume control on the video…

So you really missed the big fat label "MASTER GAIN", with the big fat blue slider bar underneath? You're obviously way too used to knobs... Back to seriousness, it was super simple to use. After I entered the serial number, I went straight to create my first preset. Everything is organized logical and so simple. You select a waveform (analog, bandlimited, continuous, arithmetic, btw) for oscillators a and b and hit "generate waveform" to get a true stereo wavetable of the mixture of the insert effects applied oscillators a and b. You can modulate through the table on a per channel basis (for example, modulating the table for the left channel, while keeping the right channel table steady). For a change of an oscillators parameters like ADSR, Filter mod or LFO, you just click on the big fat loop icon of that oscillator, which brings you to the oscillator page. The same is true for the effects (wether insert of global): clicking on them will bring up a page with all parameters to tweak to your liking. Here is an example image of an edit page for an oscillator:
https://www.soundspot.audio/wp-content/uploads/edd/2019/10/Mod-Adsr-image-2.jpg

Spogg wrote:Today I’ve been experimenting with Phase Distortion, which is a whole fascinating area to check out. We shall see! If I decide to make a PD synth it will certainly have Martin’s Chorous in it.

Can't wait to hear the results! So far I could only find our very own BobF and one other plugin programmer doing a PD synth that explore the possibilities beyond Casio!