Voice Component Modeling with the Prophet Rev2

[shiihs]

(seems like voicecomponentmodeling.com is down at the moment)

[Pym]

Yeah a lot of those ideas are in my notes to one degree or another. Once I end up exploring it more again I’ll try and communicate the process here. I’ll read through your site later this week and check out some of the specifics, sounds like you put a lot of effort into the exploration and we definitely went down very similar paths!

I’m exploring a one knob ‘seeded’ randomness per voice where it controls the range of each parameter that is then changed using the single slop knob. So maybe tuning is 0-+6 cents, vca attack is -1-+2 or whatever. But it starts at 0 for all voices. As you turn up the slop it moves them up to those values, scaling. The beauty being if you select the same random seed (maybe 0-127 or something) you can reproduce the slop behavior, and each voice number modified the seed the same way, so each voice is different but can be recalled with that seed value

Although I see the value given in full control over the offsets, this feels a little more random and ‘analog’ to me, but repeatable since you can save and recall the slop offsets

Just one of the ideas I had related to the topic

Just FYI, I have explored this concept quite extensively from many perspectives and I already have most of the code figured out for future products but it isn’t easy to port back due to design choices. I’ve always been fascinated by chaos theory and how to both bound and seed randomness in a way that emulates the part of analog we all love

I may even have some tricks up my sleeve you haven’t thought of yet... stay tuned 

Awesome, Chris!   I'm excited to see your sleeves!   This method of defining per-voice variance with a sort of "lookup table" definitely adds some classic analog realism over random slop and more artificial detuning motion.   The key is having fairly stable tuning per voice / per oscillator, but with unique offsets for each.   And tackling other parameters in the VCO, VCF and VCA sections with per-voice offsets can really create unique analog character. 

I went out a few weeks ago and captured samples of a bunch of synths and I have data tables of voice-by-voice tuning offsets for MemoryMoog, OBX, OBXA, OB SEM, CS80, Korg PolySix, Jupiter 4, Jupiter 8, Prophet 5 and Prophet 10... if you're interested.   I will say, the predominant characteristic is that all of them have these tuning offsets, and most have significant intonation variance up and down the keyboard.. I've been playing with note number modulater to replicate it.

The handling of voice-allocation would be a good area to develop further, as there are a few distinct variants (round robin, reset to voice one, backtracking voice allocation, per key voice allocation, etc..)   Currently the round robin approach is the only one available through the Rev2 modulation. 
 
It's a testament to the design of the Rev2 that it works so well even with current implementation...  but yeah, if we just had "another instance" of the gated sequencer dedicated to being a lookup table for this type of per-voice modulation offsets, I think it will be a really interesting feature on a future board.   If it had a few more lanes of data (6 or 8 lanes), and ability to capture the different voice allocation schemes, that would be awesome!   

Also, I was thinking it would be gravy if the lookup tables could be saved independent of presets... so you could just have prebuilt voice templates for a variety of different classic synths, along with variants of each where they have less-or-more voice character.   Then you could load a patch, or an INIT patch, and then navigate through a list of voice modeled presets (Voice Modeling OFF | MemoryMoog Prime | MemoryMoog Hot and Humid | Prophet 5 | etc...... )

At the bottom of the VoiceComponentModeling.com web article, in Section 5, I outlined a bunch of "future wish list" items.
http://www.voicecomponentmodeling.com/vcm.aspx 

Excited to see what comes next!

Cheers! - Jason

[creativespiral]

I've been thinking about this a lot...   I would keep the general model of the Gated Sequencer as a Voice Lookup Table:  (over a more flat table of values to all defined attributes)    Although, the idea of having a dedicated potentiometer/interface on the front of the synth to quickly toggle through saved voice templates would be totally awesome! 

Reasons why keeping the gated sequencer / lookup table paradigm is good:

1. You've already got the majority of the programming infrastructure built for it, and its a perfect setup for voice modeling.  The ideal solution may just be to create "another instance" of the gated sequencer lookup table, dedicated to voice modeling, with savable templates.   You can copy a lot of existing IP / code blocks for the gated sequencer implementation, and just make adjustments/upgrades as you see fit. 

2. The gated sequencer type of lookup table is VERY FLEXIBLE for voice modeling- Every type of classic synthesizer, or acoustic/analog ensemble that you are modeling will have different aspects to their voice-by-voice variations.   It's important to be able to address each modeled instrument differently.   ie:  some will have large variance of osc tuning per voice, while others will have very little or none, or tuning may be based on intonation vs static voice based.   

Some instruments may require more pronounced variance to attack/delay/release speeds, or only certain stages of envelope speeds.   Some will require addressing totally different characteristics of VCO, VCF, VCA sections, while leaving other ones alone.   Having ability to control each parameter independently is key if you really want to be able to capture the sound of specific classic synths, or ensembles of real world acoustic instrument sounds.   It's not a one-size-fits-all type of scenario, and the gated sequencer / lookup table model allows you to more precisely model classic synths and real world ensembles.

3. Advanced macro-type controls for voice modeling:  The gated sequencer paradigm for voice modeling also allows you to control more advanced behaviors on a voice-by-voice basis.   For instance, you might build a macro type behavior that uses a note number modulator, routed to Env 3 amount, and the overall amount of note number is controlled by a lane in the gated sequencer, giving voice by voice variance to an "Oscillator Pitch Settle" type of behavior that requires multiple modulations.   I have done this type of voice specific modeling in several patches I've built. 

4. The Gated Sequencer lookup table model also allows you to define a virtual voice count within the synthesizer... by setting a reset at step 7, you can create a virtual six-voice synth, with repeating patterns of voice imperfection.   This is very realistic if you wanted to model a classic synth from the past with a specific voice count that cycles.   You can create any number of "virtual voices" in this manner, to replicate any synth from the past, or specific ensemble sizes for voice offsets.

Below, I've mocked up a hypothetical Voice Modeling table... most of the functionality has already been developed by you guys.  The idea would be to just create "another instance" of the gated sequencer that is dedicated to allowing you to assign a voice model to a given patch... and adding some additional features to optimize it for voice-by-voice modeling. 

a. change data range:   use -62 to +62, instead of 0-125 ... works better to have 0 as a centerpoint with negative and positive offsets.

b. add a few new seq modes:   (Key Step, Reset to 1), (Key Step, Backtrack), (Assigned Key Allocation) -- this will allow you to capture other voice allocation schemes besides round robin, so you can accurately model old Prophets, Jupiters or other synths that use different voice allocation schemes..  Also, this will allow for more accurate voice modeling of real world ensembles (string sections with a principal/lead player)

c. add the ability to save the voice modeling templates / presets...  a hypothetical new board might ship with 64 factory defined voice models or something, but let users add their own user voice models if they want, or make adjustments and save them.   Would allow players to easily take a preset, or Init patch, and select a voice model scheme and build a patch from it.   If you had a dedicated knob like you were saying, you could quickly and easily scroll through voice models.   Once you save a patch, it saves the voice model with the patch.

d. maybe add a couple more lanes (six or eight) for data.   Note, even if this was not implemented as a separate instance of the gated sequencer, you could just build out the feature set on the gated seq, and add a couple extra lanes there. 

Below are some visual mockups:
Shown in the context of the Soundtower Editor style, as its a good visual representation of the capabilities of the gated sequencer paradigm.




Here's a mockup of opening a window of Saved Voice Model Templates:
A potential future board might ship with 64 factory templates of various voice models for classic synths and acoustic ensemble types.



Also, added this to Section 5 of the VoiceComponentModeling.com website:
http://www.voicecomponentmodeling.com

btw: thanks shiihs for the note about the site being down... was in process of changing servers. 

[psionic11]

Some instruments may require more pronounced variance to attack/delay/release speeds, or only certain stages of envelope speeds.   Some will require addressing totally different characteristics of VCO, VCF, VCA sections, while leaving other ones alone.   Having ability to control each parameter independently is key if you really want to be able to capture the sound of specific classic synths, or ensembles of real world acoustic instrument sounds.   It's not a one-size-fits-all type of scenario, and the gated sequencer / lookup table model allows you to more precisely model classic synths and real world ensembles.

Regarding paragraphs 1 and 2 in your post, you did something similar with the free voice-modeled templates you provided a few months ago.  Without the ability to save parts of a program (FX presets, envelope presets, voice-component gated sequencer presets), then just making separate programs, each dedicated to modelling a specific classic synth voice structure, is probably the best approach.

Regarding the quoted part above: 

As you've hinted at, there are various specific characteristics of a signature sound that determine its uniqueness.  A classic synth's voice count and tuning anomalies is one of those aspects.  Another is the particular curve (exp, log, lin) of each envelope stage, as I've brought up a few times.  Yet another is the pitch envelope -- initial flat attack (10-15ms) is brassy, initial sharp attack (5-10ms) is plucky. 

The Moog One even has an initial "Hold" stage, which is pre-attack.  Apparently, in their analysis of various classics, some keyboards have envelopes that fully open for the few ms or so before the rest of the ADSR stages even kick in.  They recommend setting this to around 10ms for emulating the "bite" some synths have.

And I want to mention phase.  Humans are binaural, and use very slight phase differentiations in arriving sounds to locate the sounds in 3d space.  Essentially a single point source of sound is reflected or refracted or received multiple times, and therefore split and time-delayed, resulting in phase differences when re-combined.  With synthesis and sound processing, modern electronic music has co-opted these naturally occurring phenomena, and so we have stereo panning, doppler effect, flanging, chorus, reverb, Leslie rotors, and a multitude of other time-based effects.

The Rev2 has positive contributions such as Pan Spread, Pitch envelopes via mods, limited envelope curve modulation, and of course its built-in FX.  The Rev2 also has a negative contribution with its single knob OSC mixer; there is some kind of phase cancellation going on as the knob approaches noon.  This is why I often choose to use only one oscillator, in order to avoid this weakening and the resulting generic-ifying result when using 2 OSCs.

How this is all relevant to your pursuits with the gated sequencer:

Almost all aspects of the characteristics I've described above have to do with the initial attack of the sound.  That's right... Initial Attack determines an overwhelming part of the timbre or unique characteristic of a sound.  And those milliseconds in the beginning, whether it's pitch or volume or phase, are a very significant part of what we register as unique, and therefore to what to devote more brain processing power, in order to identify, predict, and categorize these aural perceptions.  This is why we can tell a contrived performance using a perfectly sampled violin sound is fake, because of the lack minutiae of changing attack qualities between notes found in a live violin performance -- legato vs staccato vs accent vs vibrato, etc.

In a nutshell, one could use the stepped, gated sequencer to capture and modulate these characteristics - pitch envelope as well per voice pitch variance, envelope curve modulation, OSC waveshape.  At least in theory.

(I blame caffeine for the rant.  I should have used the morning brain power for network engineering study, but hey, I'm on holiday..)

[creativespiral]

The Moog One even has an initial "Hold" stage, which is pre-attack.  Apparently, in their analysis of various classics, some keyboards have envelopes that fully open for the few ms or so before the rest of the ADSR stages even kick in.  They recommend setting this to around 10ms for emulating the "bite" some synths have.

Yeah, a Hold Stage would be a nice addition for a future synth (in between Attack and Decay)... Rev2 has Pre-Attack Delay, which is nice for more evolving sounds... but I'd love to see the Hold stage added as well, or shapeable Attack/Decay curves can approximate the same.

Re: Binaural.... The Rev2 has positive contributions such as Pan Spread, Pitch envelopes via mods, limited envelope curve modulation, and of course its built-in FX.  The Rev2 also has a negative contribution with its single knob OSC mixer; there is some kind of phase cancellation going on as the knob approaches noon.  This is why I often choose to use only one oscillator, in order to avoid this weakening and the resulting generic-ifying result when using 2 OSCs.

It think it's not so much phase cancellation, but a factor of the way the Osc Mixer reduces volume of each osc...  I have wondered about this before.. whether the OscMixer could be adjusted to keep volume a bit higher when mixed 50/50 (adjust the curve of the mix).   Would be nice if the perceived volume stayed the same throughout the swing of the mix knob.  Not sure if the current chips can do that?

On a more general note, in regards to binaural/stereo capabilities... the fact that the Rev2 is bi-timbral really opens up so many possibilities.   In fact, I use the second layer to create stereo patches as much or more than creating classic multi-timbral stacks.  The Rev2 is capable of delivering a wide, detailed stereo field by stacking two almost-identical layers with Fixed Pan, either fully L/R layers, or a mix... you can also modulate pan spread in stereo stacks for further binaural goodness.   I've also played around with a few variations of the Leslie effect using dual layers... I have a couple patches in the bank I'm working on where I'm pseudo modeling the horn and drum speeds independently, using dual layers.   

Almost all aspects of the characteristics I've described above have to do with the initial attack of the sound.  That's right... Initial Attack determines an overwhelming part of the timbre or unique characteristic of a sound.  And those milliseconds in the beginning, whether it's pitch or volume or phase, are a very significant part of what we register as unique, and therefore to what to devote more brain processing power, in order to identify, predict, and categorize these aural perceptions.  This is why we can tell a contrived performance using a perfectly sampled violin sound is fake, because of the lack minutiae of changing attack qualities between notes found in a live violin performance -- legato vs staccato vs accent vs vibrato, etc.

In a nutshell, one could use the stepped, gated sequencer to capture and modulate these characteristics - pitch envelope as well per voice pitch variance, envelope curve modulation, OSC waveshape.  At least in theory.

I totally agree that the Initial Attack / Transient Shaping is one of the most important aspects of modeling classic synths, as well as acoustic analog instruments.   That was what I was describing in terms of more macro-type behavior modeling using Env3. 

I was able to capture measurements from a bunch of classic synths and most of them have "oscillator pitch settle" on a per voice basis.. (where there is a subtle transient effect or more pronounced pitch shift right at the beginning of the attack phase) -- its one of the characteristics I've modeled into many of the patches in the bank I'm working on... the nice thing about having a flexible voice modeling table (like the gated seq paradigm), is that you can have it varied on a voice by voice basis... the amount of these effects are not the same for each voice, and will have variance.   

I use Env3 for modeling these type of transient effects or pitch settle effects... Targeting fine tune osc pitch, filter audio mod, recursive shaping, noise mix level, osc shape, and sometimes other characteristics, depending on what instrument I'm modeling... and then control the Env3 amount through a lane in the gated sequencer (voice modeling table), sometimes scaled through mod matrix or sub modulated by note num... which gives each voice some slight variation to its transient phase.   

Also, in regards to real world instruments:  when modeling a ensemble of violins, every "player" in that virtual ensemble will have slight differences in their attack with the bow (attack speed and transient effects), the decay time, the vibrato amount, etc...   All of these aspects can be captured with a voice modeling table... resulting in much more authentic, rich sounds, especially noticeable when playing chords.

[psionic11]

Yes, there are a lot of pratical uses for gated sequence modulation of attack characteristics.

It's like the injected variation in Voice Component Modeling is the antidote to the static and precise nature of DCOs and digital envelopes.

[Gerry Havinga]

Well VCM is definitely working for me, created a first patch for the Rev2. The depth and massiveness is really nice. My first try is here (Rev2 is bass), Prophet X plays Mellotron patch:

https://soundcloud.com/user-252754541/live011-mellotron

Thanks creativespiral, awesome research and really smart thinking 

[creativespiral]

Well VCM is definitely working for me, created a first patch for the Rev2. The depth and massiveness is really nice. My first try is here (Rev2 is bass), Prophet X plays Mellotron patch:

https://soundcloud.com/user-252754541/live011-mellotron

Thanks creativespiral, awesome research and really smart thinking 

Thanks Gerry.

Nice track!.. Cranked it up and was transported to some deep space exploration    

[Naboo]

Thanks @creativespiral for sharing the knowledge.  One of the first things I do when creating a patch (on any synth) is to add random Lfo to pitch.  I haven’t been enjoying the slop on the rev2, ob6 and As-1 as much as I expected.  Reading this info and listening to some audio, I think this will be hitting that sweet spot I was looking for. Look forward to trying this out!

[SporeSynths]

Really interesting information here. Read through the article and watched the videos. Can't wait to try this out on my Rev2.

One thing I'm wondering, and maybe you've explained this and I missed it. In the video you show how your modulation method makes the upper harmonics jitter more like they do on the VCO synths. But why is it that the fundamental doesn't jitter at all with this method, since you are modulating the frequency. Is it just that the amount that you're modulating is so small that it effects the upper harmonics first before effecting the fundamental if the value was increased?

[creativespiral]

Hey SporeSynths - sorry for the delay on response.  In the VCO harmonic jitter tests (separate discussion from Voice Modeling topic), the fundamental does experience the frequency modulation, its just that the amount is very small (ie: usually less than one cent swing, sometimes a bit higher).    That swing in the fundamental seems to have a larger downstream effect on the harmonic series though.

[creativespiral]

The Voice Component Modeled (VCM) patch bank for Prophet Rev2 is now available!.. been working on this for the past six months.   128 new patches with lots of classic analog synth sounds, acoustic instrument models, VCM templates and more.   

Here's a video with some short demos of about twenty of the patches:

https://www.youtube.com/watch?v=WleQ2KrN8HY


More Info on VCM
You can read up more on Voice Component Modeling topic on:
http://www.VoiceComponentModeling.com

Some Free Examples
I uploaded some free example patches on:
https://www.PresetPatch.com/user/CreativeSpiral

Purchasing
If you'd like to purchase the patch bank, it's available here:
https://sellfy.com/sounddesign/


- Jason

[ouzoman]

Great set of patches!  Congrats!

[Gerry Havinga]

Completely blown away by @creativespiral's patches. Beautiful lush sounding, very organic and massive. The VCM technique really works best with chords and layering A+B, I will have to invest in the 8 voice upgrade and getting my Rev2 up to full strength   

@creativespiral if you are ever in the Netherlands, I do owe you a meal and a beer, thanks!

[gernotreininger]

Hi creativespiral. I just purchased your great set of patches. You did a fantastic job and I guess a can learn a lot from your work. Thank you for your creativity and keep on your awesome work.

[creativespiral]

@ouzoman, @gerno, @gerry 

Wow...  Thanks guys!!  Appreciate the feedback.   

btw, if you don't already have an expression pedal, I highly recommend picking one up... there are modulation pedal mappings on a lot of the patches...  some have significant morphing of sounds.   

[Darrolav]

Hi creativespiral 
I bought your patches and I have to say that your attempt on vangelis cs80 sound is just amazing
Thank you so much for your work

[gerardplouf]

Hello Jason,
Just buy your patches. You make an incredible job.
Thanks for these great soundbank.

[Tugdual]

creativespiral you have developped interesting concepts but I believe the real breakthrough would be to access calibration data. I remember the Prophet generating crazy sounds out of calibration, looks like it is far from having delivered all its potential.
Another dream that I have is sequential deciding to share the firmware as an open source.
Somehow Korg started that with their custom generators.

[creativespiral]

@geradphouf, @Darrolav - glad you're enjoying the bank - thanks for the feedback guys!  Appreciate it!

@Tugdual - Yeah, the Voice Modeling is basically that type of approach (modeling calibration offsets for tuning, envelope behavior, noise levels, and various other characteristics, on a voice-by-voice basis)   

Once you set up the voice modeling data table in gated sequencer, you can dial in classic analog type intonation tuning offsets (where each oscillator in each voice has increasing or decreasing offsets to tuning, tied to position on the keyboard)... this is the most common behavior I measured in my tests of classic VCO synths (I tested OBX, OBXA, CS80, PolySix, MemoryMoog, Jupiter 4, Jupiter 8, P5v3, P10, and some others)

Or with a little extra adjustment of the data table, you can dial in crazy, de-calibrated old synth character... I've got a couple patches where I push the limits of what sounds good, trying to dial in a sort of classic "MemoryMoog on a hot and humid day" sound, where there's a ton of "de-calibrated" behavior in regards to voice tuning offsets and envelopes.   It works better for certain types of playing/music... If you're playing synth power chords (1,5,8) and octaves you can really push the decalibration, and it can sound really good with lots of classic/organic flavor.   If you're playing more complex passages, then the more dialed back approach works best.   The beauty of it is that once you setup the modulation wiring and lookup table in gated sequencer, all you have to do is swing some values in the seq to make it more wild, or more tamed.

There's some patch examples that you can download for free here if you want to try them out:
https://www.presetpatch.com/user/CreativeSpiral