Voice Component Modeling with the Prophet Rev2

[creativespiral]

I would love to hear some comparisons some day!

I wonder if you would give a more detailed run down of your templates some time. For example, do they all have similar amounts of jitter? Would one be better for monophonic patches? What is the difference between patches with different tuning settings (-4 vs. -32 if I remember right). Thanks!

I'll try to not turn this into a book... a few notes that are sort of big picture items:

1. Modeling VCO Jitter / Harmonic Jitter with a DCO or digital oscillator -- I usually only use this modeling technique on simpler sound designs... mono basses, leads, and simpler poly sounds.   The effect is subtle - I think it's basically drown out with any sort of more complex patch design with other LFO modulation going on, filter modulation, or lots of oscillators going at once.   It takes 1-2 LFOs to pull off the effect realistically, and to me LFOs are at a premium value for sound design... this decision to ignore the jitter on many patches comes down to not wanting to sacrifice the LFOs and just feeling like in the grand scheme, I don't notice the difference as much for my completed patches.  The audible effect of this is nowhere near as significant as the Per Voice modeling for component values.

2. There's two main approaches to modeling oscillator tuning - either 1. assigning a per voice / per oscillator offset to each, and that offset will be the same up and down the keybed whenever that voice is triggered.   Or 2. multiplexing in a note number modulation slot to give each oscillator an offset, which also includes the intonation / osc scaling tendency over multiple octaves.   In the great majority of synths I tested, the latter approach is what I measured... synths tend to have tuning scaling issues over a five octave range where they will go either flat or sharp progressively as you go up/down the keybed.   Sometimes the tuning varies just 3-8 cents over five octaves on a given voice, and sometimes significantly more.  I tested two MemoryMoogs, and they had particularly bad performance (lots of character) over five octave range (some oscillators having around 15-20 cents variance over that range).   So, for max realism I recommend the intonation/notenum approach most of the time.. In the Rev2, it does take up several mod slots (4-5 total) to do this, whereas the straight voice offsets just use 2 mod slots.   The straight per-voice, per-osc modulation gets you 90% of the way there though in terms of capturing the tuning voice modeling character... but yeah, if you really want top realism, spend the extra two slots to model it with intonation too.   I use both methods.

3. After setting up modeling for Osc1/2 tuning per voice, the next highest priority items for me are usually giving offsets to Env Attack and Env Decay per voice.  If you add just small offsets to these parameters on a per voice basis, you'll capture the slight timing variation that classic envelopes have voice to voice... and the downstream effect is that the filter cutoff point will have temporal offsets from voice to voice, so you're sort of killing two birds with one stone.  This works great for big stacked or unison Sawyer type of filter sweeps.   You can model any number of other parameters with voice modeling too... some patches I directly target filter cutoff per voice - others I will model a macro pitch settle effect at the transient attack and have that vary per voice (this is a common effect of old synths, to have a quick transient pitch settle at the start of each new note).   And there's many other variations that can be targeted per voice, just depending on the specific sound design you're working on.

[Jason]

Thanks for the detailed answers!

If I wanted to see if a sound has the jitter effect, what can I look for?
I assume that templates like: #117 VCM 4V3O (Int, Jit) and #120 VCM 5V2O (Int, Jit) have it based on the Jit of the title?

Two of the best Oberheim sounds that I've ever heard on any Sequential gear are:
#40 BRS 80s Classic OB2T (What does the 2T refer to?) and #44 BRS AnalogBrass VCM. (These are especially big given how I am panning two Rev2s Left and Right.)
On these patches, the fine tuning is -31, while the fine tuning of your templates (e.g., 117, 120) is +4...
Subdivisions OBX fine tuning is +21...
SYN GX1 JPJ Love2 fine tuning is -7
Would you please explain why the differences?

Similarly, why does one of the Adam Holzman Leads have a fine tuning of -31, while the other is -2?

Many thanks again,
- Jason

[creativespiral]

Thanks for the detailed answers!

If I wanted to see if a sound has the jitter effect, what can I look for?
I assume that templates like: #117 VCM 4V3O (Int, Jit) and #120 VCM 5V2O (Int, Jit) have it based on the Jit of the title?

Two of the best Oberheim sounds that I've ever heard on any Sequential gear are:
#40 BRS 80s Classic OB2T (What does the 2T refer to?) and #44 BRS AnalogBrass VCM. (These are especially big given how I am panning two Rev2s Left and Right.)
On these patches, the fine tuning is -31, while the fine tuning of your templates (e.g., 117, 120) is +4...
Subdivisions OBX fine tuning is +21...
SYN GX1 JPJ Love2 fine tuning is -7
Would you please explain why the differences?

Similarly, why does one of the Adam Holzman Leads have a fine tuning of -31, while the other is -2?

Many thanks again,
- Jason

Thanks!  Yes, those templates with "Jit" have jitter in them.   As mentioned, I don't tend to model the jitter in a lot of patches, just because of the multi-LFO requirement to do it accurately and I just don't hear that much of difference in more complex sound designs...  The effect of the Voice Modeling the Tuning Offsets and other Env/Filter characteristics has a much more dramatic effect than Jitter, and that's where I spent most of my time in research and sound design / modeling.   There are a few patches in the soundset that have it modeled with Jitter though.  I used the term "VCO" in the name of a few, to indicate they have VCO jitter modeled...  You can model the jitter with a single triangle LFO to get "in the ball park", or use an additional second random LFO as well to get closer to what I measured in various VCOs... and this can also be modeled per voice as well (and even if not modeling it with separate per-voice values, you still need to use the gated seq and mod matrix to scale down the values to appropriate ranges)   For VCO jitter modeling, I usually use a rate around 54-60 for the LFOs, just based on what I found.


The patches like those two BRS ones you mention, where you see fine tuning as -31 value, are patches that have standard voice tuning offsets to each oscillator (without intonation / osc scaling modeled).  Whenever you hit the associated "virtual voice" based on how many steps you use in the mod sequencer, the offset for that voice/osc will be the same.    The -31 to fine tuning is just an offset so that the voice modeling via the gated/mod sequencer zeroes out the tuning at a value of around +62, which is about the centerpoint of the mod sequencer value range.   It's just a counterbalance, so there's room to tune both flat and sharp per voice in the sequencer/lookup table.   I think the majority of patches I've developed as standard tuning offsets per voice use this '-31' offset value... that became my standard template.

The #54 Subdivisions OBX patch uses the intonation based per-voice offsets (aka osc scaling per voice)...   you'll notice in the mod matrix, there are four slots used for voice modeling (1&2 modulate the amounts of 5&6, which use note-number to achieve oscillator scaling.   For this patch, along with some others, I used measurements I took from an Oberheim OBX synth as references, and tried to get as close as I could to the actual tuning tables I cataloged.   One of the common behaviors I measured in several classic polys is this osc scaling / intonation, and specifically where the oscillators tend to be more sharp in the lower octaves and progressively get more flat as you go up several octaves.  So that was how I did the voice modeling for this patch and some other OB style ones.   In this case, the +21 was required as an offset to make it so that the center of the keybed is mostly in tune for most voices (around C4/C5)  The exact tuning is of course different per voice/per osc, but with this intonation effect modeled.

The GX1 John Paul Jones All of My Love patch uses the per-voice intonation modeling also, but instead of oscillators getting more flat as they progress up the keybed, they are getting progressively sharper.  The smaller '-7' offset to tuning on this patch is just balancing the way that note number is scaled up from C0 on the Rev2.   That value was used to get the keyboard in most nominal tuning in the C2-C4 range for that patch.   The offsets for the intonation based patches just adjust which area of the keybed is "most in tune", which is how classic synths behave..  They usually have an area somewhere between C2 and C5 where they are pretty well in tune, but as you diverge from that sweet spot, things get progressively looser in tuning, and you get a lot more vintage character. 

For the two Adam Holzman Lead patches, the #98 patch was modeled from a patch that I use constantly on my Poly Evolver Keyboard, but added just a bit of voice modeling to it and the VCO jitter modeling... That's one of my favorite lead sounds.  The '-2' offset in that patch was all that was needed, just based on me using less of the range in the mod sequencer... I think that patch was one of the earliest voice modeling tests I did.     The #100 Holzman Lead patch is a more advanced variant on that sound design using stacked, hard panned layers, and my more standard value ranges for regular per voice modeling.

[Jason]

Thank you so much for such a detailed response! It helps me understand what’s going on be behind the curtain, as your programming skills are on another level. I think the Rev2 would be even more popular if people could hear your patches when they are demoing it.