So come up to the lab / and see what's on the slab / I see you shiver with antici


1500 posts!

Meanwhile, over at the MW site, someone by the name of MATsimile has posted this photo, possibly of a reflection in Richard Devine’s left cornea:

So, Stages segment generator. Six sliders with LEDs, six mini-pots to the top labelled Shape/Time, then six buttons that appear to change mode for each slider, indicated by a coloured LED above each slider, then maybe six more mini-pots labelled Trig…(something), presumably setting the trigger behaviour.

Clearly this is a six-segment envelope generator. Now for the speculation: with configurable shape and duration per segment, and probably other configurable behaviour (at least segment disabling, probably configurable looping behaviour, possibly inter-segment cross triggering). It may be that the segments can be divided into groups, so you could have, say, three AD envelopes? And of course, overall looping so it can act as an LFO. And voltage control over some things? So a sort of Rossum Control Forge Lite, but with physical controls rather than a complicated conditional programming model, and maybe some similarities with the ill-fated OmniMod module? Now, is it just one bank of six segments, or six independent channels each of six segments? The latter seems more likely, although that complicates the UI and the physical control-to-value mapping (like in the expert mode in Peaks), so possibly not.

Of course it may do flat segments as well in a clocked mode, in which case it’s also a six-stage sequencer. Quantised? Probably not, expect a separate quantiser, I suspect.

Another new module is also visible to the left of it, with labels of Spread and Bias.


Picture is from Marcus Fischer’ instagramm

Definitely the october '15 facebook teaser module


You sleuths make this waiting so much fun! I feel like I’m reading a Hardy Boys mystery back when I was 10 - more clues, more excitement!
Certainly NOT helping me break any internet addiction I might be struggling with…

The scary question for me is: I have only 16HP free in my rig!
Hard choices ahead, I fear…


I was being flippant, channelling M.C. Escher, but yes, Marcus Fischer is the source of the leak. Here is a better shot from Instagram, showing more detail:


Define “more detail”… :wink:



He means “more giant polygons of light” like a Brian Eno installation.


Yeah, by some definition of more detail…

This is better:

It shows a symmetrical two-channel layout, with labels Bias (so, mean), Spread (so, variance), Jitter (time-variance), Clock, ?Gate. Something that samples from a PDF (probability density function)?


Given Oliviers recent design choices, i highly doubt it will be both multi channel and six stage.
He explicitly stated at many points that he wants to move away from hidden functions and paging interfaces.
My guess is that this will be six channels of voltages that can either be chained for complex shapes or used individually for utility type lfos/envelopes. This way, paging could be avoided while keeping it flexible and universal.
Might also be able to do 3 2-stage or 2 3-stage envelopes?

That said, the module to the left of it looks very much like some sort of complex multichannel random to me. I could imagine jitter being a variable “range” of randomness while bias mighr move from more periodic shapes to increasingly chaotic behavior?

In any case im tremendously excited for these.


Interesting speculation.

So, presumably two different modules.

One with sliders for 6 “segments” and another, primarily with all knobs, but with six outputs.

This 6/6 relationship is intriguing…or perhaps means nothing at all.



I imagine it more like a set of 6 arbitrary steps that can be rising, falling or static (adjusted by the knob) with adjustable time (the slider most likely)
Then, you decide what you want to happen, should they fire in sequence, independently from each other, chain the first three and the last three and so on.
There seems to be a knob on the bottom as well? Maybe an overall time/clock div. Adjustment?


Well, the Olivierian (Braids/Peaks) envelope framework requires several parameters for each segment:

  • Starting level
  • Wavetable (determine the segment shape)
  • Target level
  • Phase increment (smaller increments give longer segment time)
  • Next segment

Starting level is usually set to zero for the first segment, and to whatever the target level is for the previous segment for all subsequent segments.

Target level is usually set to the maximum value for the attack segment in an ADSR envelope, and to the sustain level for the decay segment in such an envelope. The target level for the final segment (decay segment in AD envelopes, release segment in ADSR, ADR or ASR envelopes) is usually set to zero. But for multi-segment, more general envelopes, you probably want to be able to set the target level independently for each segment except the last segment. In fact, it isn’t even necessary for the first segment to start at zero or the last segment to end at zero (have a target of zero), but it is desirable for both those levels to be the same (so it ends at the level at which it starts and v-v).

Next segment number is usually just current segment + 1, unless looping across subsets of segments is desired when the gate is held high.

So the design problem is how to map those parameters onto a set of pots (looks like 12 rotary pots and six sliders for six segments in the new Stages module), plus 6 (or more) buttons. It will be interesting to see how Olivier has tackled this challenge (guaranteed to be tackled well), if all these surmises turn out to be true.


ohh yeah



Commence tizzy…


Marbles … maybe even today!


maybe we get a easter gift :sweat_smile:


In lieu of built-in Easter Eggs, yeah…I’ll take it!



Hey, who knew that Olivier, like Darwin, had been on a field trip to Oz to gather inspiration for this module - as he says in a comment on that Instagram post (and no capybaras, hey, I’m verbatim quoting here):

il s’agirait plutôt de deux ornithorynques de petit taille.

Anyway, we don’t see much more in the Instagram clip: we have clock in, and bias and spread controls (and CV for them). But we now know it outputs CV, specifically pitch CVs, probably quantised. Two channels. I think this is stochastic melody generator (d’uh!), with probability of each note being chosen according to a PDF (probability density function). The button and LED at the top might allow choice of distribution. Bias sets the mean, spread the variance of the distribution (but that means a redundant control for the Poisson distribution…). I dimly recall that I mooted such a thing years ago?

But I suspect there is more to it. It might implement a machine learning algorithm to learn a generative model for the melody. If that Pugtronix Oracle module (is that what it was called?) hadn’t appeared a year or two ago, I’d wager it used factor oracles, but Olivier hates to do things that have already been done.

If it really is called Marbles, it might be some variation on a Galtonian quincunx? Like a pachinko machine. A marble falls through a lattice of pegs. Each peg is a note? But the melody sounded fixed for any given setting of bias in the Instagram clip. Hmmm, thinks…

Simple stochastic notes, even if selected according to probabilities from an underlying distribution, seems a bit unsatisfying, even if it the first module to offer random notes from anything other than a uniform distribution - it must be higher-order than that.

Edit: outputs labelled X~2~ and X~3~ are visible. Polynomials?


I don’t speak French, but I did see that that quote was in response to a commenter who said “Is that a new module?” and I took the response to be a sarcastic “no, it’s actually two platypuses”. Hardly Oz if so :stuck_out_tongue_winking_eye:


Marbles could also refer to two or more interleaved or related sequences, as something which is ‘marbled’ is not uniform but patterned


Platypuses (strictly, platypodes) in the wild are vanishingly rare outside of Oz. And they are rare enough here.

But it is clearly a reference to the use of monotremic gradient descent algorithms to find local nadirs on a high-dimensionality anti-melodic hypersurface.