You are here

Behringer DeepMind 12

Polyphonic Synthesizer
Published December 2016
By Gordon Reid

Behringer DeepMind 12

Behringer’s first foray into the world of synthesis always promised to be interesting, and the results do not disappoint...

Behringer’s reputation in our industry may not have always been the greatest, but in recent years, Uli Behringer’s creation of The Music Group in the Philippines to act as an umbrella organisation for the Behringer brand and other acquired companies (including Klark-Teknik, Lab Gruppen, Lake, Midas, Tannoy, TC and Turbosound) has done much to improve the company’s standing, although there are still products being released that exhibit, um... the sincerest form of flattery with respect to products from elsewhere. In the synthesizer arena, Uli Behringer has long expressed an interest in cloning synths such as the Minimoog, the Prophet 5, the Odyssey and, in particular, the Juno 106. (Google the Phat 108 if you’re unaware of this.) So, when teasers regarding the DeepMind 12 started to appear, it wasn’t surprising to see that it owed much to the Juno. The similarity of graphic design was remarkable, and even the choice of the Juno’s fader caps reinforced the feeling that there was going to be something uncomfortably Roland-esque about it.

When further details emerged — the use of DCOs, the switchable bass boost to emulate the underlying frequency response of the Junos, the filter topology and more — it was clear that the similarities were going to be more than skin deep. But it also became clear that there would be significant differences and, when the DeepMind 12 emerged, it had grown a second oscillator, a dual-mode filter, a modulation matrix, effects and more. So while the inspiration certainly came from across the East China Sea, Behringer’s new synth is not a knock-off of the Juno; in part an homage, maybe, but not a clone.


Much of the design and implementation of the DeepMind 12 comes from the Midas team in the UK, which bodes well since they have decades of experience in the higher end of the audio industry. But as far as I’m aware, this team hasn’t designed a synth before, so I expected some quirks, and I wasn’t disappointed.

Let’s start with the basics. Despite being described widely (and not least by Behringer themselves) as ‘analogue’, the DeepMind 12 is a mono-timbral 12-voice analogue/digital hybrid polysynth with a lightweight but useable 49-note velocity and aftertouch sensitive keyboard. Each voice comprises two DCOs, a switchable 12/24dB/oct VCF, three digitally generated contour generators, two digitally generated LFOs, and a modulation matrix. Voices can be stacked in various ways from 12-note polyphony to single-note detuned monstrosity and, following a global high-pass filter, four configurable digital effects units complete the audio path. Performance controllers are limited to the traditional pitch and modulation wheels, although there are also octave up/down switches and multi-mode, polyphonic portamento. There are 1024 patch memories in eight banks of 128 sounds, and you can allocate these to 16 categories, four of which can be user-defined for specific projects.

The first hints that the DeepMind 12 was not going to be multi-timbral (or even bi-timbral) appeared when its rear panel was revealed. This comprises just two (L-R) balanced outputs, a  stereo headphone output, a  sustain pedal input and an expression pedal input, all on quarter-inch jacks. Next to these, there’s MIDI in/out/thru on five-pin DIN sockets, and MIDI in/out on a  USB 2 type B socket. (There’s no USB audio.) Add an IEC power input and its associated on/off switch and that’s your lot. The first hints that the DeepMind 12 was not going to be multi-timbral (or even bi-timbral) appeared when its rear panel was revealed. This comprises just two (L-R) balanced outputs, a stereo headphone output, a sustain pedal input and an expression pedal input, all on quarter-inch jacks. Next to these, there’s MIDI in/out/thru on five-pin DIN sockets, and MIDI in/out on a USB 2 type B socket. (There’s no USB audio.) Add an IEC power input and its associated on/off switch and that’s your lot.

Perhaps the most surprising thing about the DeepMind 12 is its size. With a four-octave keyboard and a shallow control panel, it’s right down at the dinky end of the spectrum. Nonetheless, it feels reassuringly solid and its keys are full-sized, so that’s one hurdle safely cleared, and it has an integrated power supply, so that’s another. Unfortunately, the limited panel area means that this is no ‘one knob per parameter’ synthesizer, and many basic functions are accessible only via the on-screen menus. Moreover, the monochrome 128 x 96 screen is far from cutting-edge, and how the draft manual can call it “large” in 2016 escapes me. Apparently, there was a discussion within Behringer about replacing it with a full-colour TFT display but, for both cost and development time reasons, the company decided to stick with the existing device and, to be fair, it does the job. Staying with the hardware for a moment longer, I was rather bemused when Uli Behringer wrote to a number of forums stating, “With approx. 4000 components, the DeepMind 12 is one of the most complex synthesizers that has ever been built. Due to its discrete design structure, it has vastly more components than you will likely find in any of the current synthesizers on the market.” This seems to be a remarkable claim. The analogue board in the DeepMind 12 is tiny and stuffed to the gills with chips. Should this matter? Not in the slightest, but I find his comments to be (at best) odd. On a more positive note, he also went on to say that the DeepMind 12 will come with a three-year warranty, which is generous and welcome.

Before starting to review the synth, I investigated its five Global menus. The first handles connectivity, allowing users to set up MIDI and SysEx for its five-pin, USB and Wi-Fi connections. The second contains the keyboard settings: things such as velocity and aftertouch curves, local on/off and the basic octave. Next come the pedal and panel settings menus, the first of which provides various pedal modes and, amongst other things, a range of sustain/gating options that EDM artists are going to enjoy, and the second of which allows you to determine how the physical faders respond when you move them. Finally, there’s the System menu which, perhaps most importantly, provides access to the various tuning and calibration routines. So, having set everything up to my liking, it was now time to create some sounds...

Sound Sources, Filters & VCAs

The importance of the DeepMind 12’s menus becomes apparent as soon as you turn to the first of the building blocks of a patch (which Behringer call a Program). On the panel, DCO1 offers just on/off switches for its sawtooth and pulse/PWM waveforms, plus faders to control the pitch modulation and PWM depth. That’s not much, and its seven further parameters (including the footage and the modulation sources) are tucked away in a menu. Next to this, DCO2 offers faders for pitch, pitch-modulation and level. Surprisingly, there are no waveform switches, either physical or in the associated menu, and it transpires that this oscillator only generates a square wave. However, a fourth fader allows you to control a novel form of waveshaping, in which a pulse wave interrupts each half of the duty cycle; the higher the value, the wider this pulse becomes. Initial tones can range from nasal to metallic and, since you can modulate this parameter, DCO2 proves to be more flexible than you might expect. You can also hard sync DCO2 to DCO1 with the usual range of results, and the combination of waveshaping and sync offers some interesting timbres. The final audio source is a pink-noise generator with its own level fader. Be aware, however, that there’s no mixer on the DeepMind 12; you can only set the DCO1 waveforms to be off or on (zero percent or 100 percent) and fade DCO2 and/or noise into the mix at levels lying between these bounds, which will limit the range of sounds that you can create. Furthermore, there’s no external signal input, which may be a significant omission for some potential users.

The low-pass filter offers the expected frequency and resonance faders, with additional faders for keyboard and contour tracking as well as LFO modulation. If this sounds rather familiar, I’m not surprised because it mimics the Juno series right down to the graphic design; apart from the fact that the contour invert switch is in a different position, they are identical from any practical point of view. Nonetheless, there are two additional buttons here, and these are far from trivial additions. The first switches between a 12dB/oct (2-pole) response and a 24dB/oct (4-pole) response, and very useful it is too. In fact, I soon found that I preferred the 12dB/oct mode for many sounds. The second takes you to the filter menu, where you’ll find additional parameters such as velocity sensitivity and aftertouch LFO depth. Behringer claim a cutoff frequency range of 50Hz to 20kHz and, when tested, I found these figures to be almost spot-on.

The editor software’s Overview page.The editor software’s Overview page.

When the DeepMind 12 arrived I tried to program some of my favourite Juno sounds using the self-oscillating filters as additional oscillators, but each filter was out of tune with the others and they tracked at slightly different rates, so I ran the calibration routines. This took a while but, once completed, everything was in tune and tracked well. However, unlike the Roland’s filters, the Behringer’s don’t lock to the harmonics of the oscillators, so I was unable to recreate some of the patches I wanted. Also, as I later discovered, the tuning doesn’t always survive being switched off for a while and then switched on again. Unless this is a quirk of the pre-production review model (which is quite possible) you may find yourself running those routines more frequently than you might expect.

The next stage in the signal path for each voice is the VCA. There wouldn’t be much to say about this except that it has an unexpected pan parameter that allows you to spread the 12 voices across the stereo field. By default, this is set to zero (mono) so you may not be aware that it’s there, but a little bit of width can add interest to many sounds such as pads and special effects.

Following the VCAs, the voices are summed to stereo and passed through a global 6dB/oct high-pass filter, which is another nod in the direction of the Junos, and a useful addition on any synth. There’s even a button marked Boost which, when on, mimics the bass response exhibited by all of the synths in the original Juno series, and this does a fine job of thickening and warming up your sounds.

LFOs & Contours

To the left of the audio oscillators, you’ll find a pair of identical LFOs. These offer a choice of seven waveforms that you can only select via their menus, with faders to control their rates and delay times and a menu item to adjust their slew rate if desired. Being digitally generated, they can offer options that wouldn’t otherwise be practical, including synchronisation to (in effect, re-triggering by) the arpeggiator clock, and allowing precise control over the relative phases of the LFOs between voices. What’s more, their frequency ranges are uncommonly wide, reaching from around 0.05Hz at the bottom end to 1.28kHz at the top, and you can make them track the keyboard accurately, which means that you can use one to generate AM and FM timbres and the second for conventional modulation duties. Given the fact that the audio oscillators are analogue, the consistency of FM tones across the voices is remarkable.

I’m not going to waste space here arguing the perceived pros and cons of analogue versus digital contours — it doesn’t bother me one way or the other as long as they’re snappy at one end of the scale and languorous at the other. In this regard, the DeepMind 12 scores highly. I measured minimum AD times so short that the VCF and VCA barely had time to click, and maximum attack and release times of 32 seconds. But rather than concentrate on these measurements, I’m going to compliment the designers on a rather different but astute bit of design; each of the A, D and R stages can take the form of an exponential, linear or logarithmic curve, with 256 variations from one extreme to the other, and the sustain phase offers control over both its start and finish level. These curves allow you to imitate the characteristics of many vintage synths and, by allowing you to define shapes that would normally be the province of five-stage contours, also make it possible to program things such as orchestral sounds more accurately than would otherwise be feasible. With multiple triggering options that you can select independently for each of the three contour generators, I’m impressed. Furthermore, while two of the three contours are connected directly to the VCF and VCA, all three appear as sources in the modulation matrix, while all of their rates, levels and curves appear as destinations. This is good stuff.

Ah yes, the modulation matrix... This offers eight slots within which your choices from 22 sources can modulate (with positive or negative polarity) your choices from 130 destinations, and it’s where you do interesting things such as shorten the contours as you play up the keyboard for natural sounds, or create Programs with multiple parameters of pressure-sensitivity for all of your favourite ARP Pro-Soloist imitations. (What do you mean, you don’t have favourite Pro-Soloist sounds? Shame on you.) There are also two modulation sources here that aren’t visible anywhere else on the synth: MIDI Note Off Velocity, which can be ‘real’ or set to a fixed value in the global menus, and MIDI Breath Control, which is mapped from the expression pedal input. Of course, there’s nothing stopping you from modulating thingies that are themselves modulating other thingies, so the possibilities are enormous.

The Effects

The last stage in the signal path comprises four digital, stereo effects slots, although you can bypass these to retain an all-analogue VCO/VCF/VCA/HPF path to the outside world should you wish. There are 33 effects available and, to use Behringer’s description, the “device inspirations” of some of these were products manufactured by companies that now find themselves within the Music Group, while others from elsewhere include reverbs inspired by the Lexicon PCM70 and 480L, a Roland Dimension D chorus, and a Moog-style filter. You can choose to pass some or all of the signal through any of these, and although the 24-bit word length, 48kHz sampling rate and maximum 40-bit calculation resolution aren’t to the highest specifications offered by many of today’s processors and plug-ins, they are in my view entirely adequate here.

There are 10 different effects configurations, ranging from the obvious ‘four in a row’ to all in parallel, and eight others between, including two with feedback loops. You’ll need to use the last of these with care since it’s easy to send the whole shebang howling into the nether regions of sonic destruction. Fortunately, the designers anticipated the worst consequences of this (your speaker cones sticking out of the opposite wall of the studio) and inserted a 30Hz high-pass filter into the loops, which could save you a fair bit of money if things go awry.

The effects section may appear clearer when laid out in the software editor.The effects section may appear clearer when laid out in the software editor.

The manual (and, as I later discovered, the editor) illustrate the effects algorithms in graphic form, showing their parameters in renderings of the original processors. The synth’s screen is unable to do this, so parameters — up to 12 for a given effect — are shown as slightly square knobs with three-character shorthand names. Happily, many of these parameters are destinations in the modulation matrix, which suggests yet more possibilities. I was pleased to find that you can reorder the effects within a given Program without losing their parameter values, and that you can copy them from one Program to another. On the down side, there are no physical switches to turn the effects on or off. This is fine in the studio, but you’ll probably want to allocate controls in the modulation matrix to make this practical in live situations.

Chords, Clocks, Arpeggiation & Sequencing

The first facility you’ll encounter in the Arp/Seq section is the chord memory. As you would expect, this allows you to record a chord and transpose it up and down the keyboard as you play. You can even program different chords for different keys, which some people are going to love. But my favourite use for this was somewhat different, because I employed it to recreate some huge, modular synth patches. I did this by programming a suitable single-oscillator patch and then recording a chord (say, the first, fourth and fifth in the scale) and playing the resulting three-oscillator patch using two, three or even four voices per pitch. The results could be remarkable.

Moving on, the DeepMind 12 has an internal master clock that can drive its arpeggiator, control sequencer, and any synchronised functions such as the LFOs and some effects parameters. You can set this in the range 20 to 275 bpm using the Rate fader, or tap the tempo using the Tap/Hold button. Alternatively, if a MIDI Clock is presented, the synth will synchronise to this with clock divisions ranging from 1/2 to 1/48. Swing is also available within a range of 50 (no swing) to 75 percent.

The arpeggiator itself offers more than you might expect, not only providing the common up, down, up/down, random, and ‘as played’ modes, but five unusual ‘inverted’ and ‘alternating’ modes across a massive six octave maximum range. There are also 64 patterns available: 32 presets, and 32 that you can program yourself. These, in effect, turn the arpeggiator into a monophonic, 32-step sequencer. You can edit the patterns on the DeepMind 12 itself, choosing the pattern length, the note, and the gate duration for each step, but you can’t program the notes from the keyboard, which makes things rather laborious.

The final facility available here is the Control Sequencer which, as the name implies, doesn’t play notes but controls patch parameters via the modulation matrix. The controls are similar to those of the arpeggiator, with the addition of a slew generator that allows sequences to evolve smoothly from one value to the next rather than jumping between steps.

In Use

Thirty years ago, the Juno 106 was hugely successful not just because it was affordable but because it was an astoundingly simple synth to program and use. In contrast, the extensive use of menus on the DeepMind 12 recalls a later era. Four cursor keys are placed around a clicked knob that performs the function of the Alpha Dial introduced on Roland’s Alpha Juno series, there’s a data entry slider which speeds up large changes in values, and if you want to jump to the beginning or end of a range of values you can do so using the Roland trick of pressing and holding + followed by -, or vice-versa. In most instances the system works well, although it may come as a bit of a shock for younger players brought up on touchscreens. Furthermore, despite the diminutive size of the control panel, I never found it cramped. Compact, but not cramped. I also liked the provision of the voice LEDs at the bottom right of the panel. These proved useful when playing stacked voices (they provide immediate information about how many keys you can press simultaneously before note stealing occurs) and they also helped me to decide when a specific voice was out of kilter and I needed to run the tuning/calibration routines.

But what of the sound? In truth, there are already many demos on the web, but what these often fail to illustrate is the range of sounds that the DeepMind 12 can produce. I stumbled across high-quality patches that ranged from brittle and glassy tones to wonderful lush ensembles, pads and carpets of sound. In between lay orchestral sounds, electric pianos and clavinets, organs, and more. Delving more deeply, the arpeggiator and control sequencer can take it deep into EDM territory, where it will happily squeal, fart and go splat. I was also surprised at how good a monosynth it makes. You can go over the top with stacking and detuning voices, playing with the relative phases of the LFOs, and applying effects but, used more judiciously, these features also make it possible to obtain a huge range of useful lead and bass sounds. This is also, perhaps, where its four-octave keyboard works best. Too short for many polyphonic uses, it’s ideal for soloing, especially since you can determine the note priority (lowest, highest, last note) and whether it responds to single- or multi-triggering. Interestingly, despite this range of sounds, there’s a consistent character at the root of all of them. Just as there was a Roland sound, a Moog sound, an Oberheim sound (and so on) in the past, there’s an underlying Behringer sound in the DeepMind 12 which purists will immediately identify as emanating from something more modern than the VCO-festooned synths of yesteryear. Some will love it, others won’t, but that’s a personal choice. Objectively, I have just one criticism in this area; the pre-production unit that I have here is rather noisy. This is particularly apparent when the sound contains modulated effects, because swirly noise is much more intrusive than consistent hiss, so I queried this with the engineers at Behringer. They told me that the production models will be at least 6dB quieter than this one, which is good news, although I suspect that they could do with going somewhat lower still.

To my surprise, I don’t have much else to criticise. Sure, DCO2 is strangely limited, the lack of a mixer is a pain, the inability to select certain functions from the front panel is frustrating, editing can be a bit clunky, and there’s no analogue connectivity. But, at the price, I think that I can accept this. On the other hand, building a performance-oriented polysynth with a four-octave keyboard has always seemed a bit tight-fisted. I’m also disappointed that the DeepMind 12 lacks a bi-timbral mode because, in some ways, it’s reminiscent of the mighty Roland Super-JX10, which allowed you to position two patches either side of a split point on its capacious 76-note keyboard, or layer them to create sounds with two distinct sonic components. It would have been great had the DeepMind 12 been able to do the same.

Finally, I have to mention the DeepMind 12’s cooling fans. I own a number of synths with very noisy fans and, while they’re acceptable on stage, I try to avoid using them in the studio whenever possible. I was therefore surprised to find that a small synth based on the latest surface-mount technology should require not one but two fans, which suggests that there’s a lot of heat being generated by its XMOS and SHARC processors within an enclosure that’s too small to dissipate it using simple heat sinks. Knowing how tightly Behringer control costs, there’s no way that these fans would have been added if unnecessary, so it’s just as well that they generate less noise than the monsters in, for example, my Memorymoog and some vintage Kurzweils. Whether they will remain quiet in years to come is something that only time will tell. Strangely, you can switch the fans off in the Global menus. This seems odd to me — either they’re necessary or they’re not — and I was tempted to switch them off for the review to see what happened. I suspect that nothing worse would have ensued than everything going out of tune but, since the chaps at Behringer undoubtedly want the synth back in a functional condition, I decided not to take the risk. However, I did slow the fans down to reduce the acoustic noise because one of them was plainly audible if I failed to do so.


Knowing Behringer’s history, I must admit that I came to this review with a degree of trepidation. But as I spent time with the DeepMind 12, I warmed to it. Sure, it’s not the analogue synth of your wildest dreams, but it was never designed to be and, with a projected price of £799 to £899 in the UK, it will cost around one third of the price of flagship synths such as the Prophet 6 and OB-6. Or, to put it another way, it will cost (in real terms) a fraction of the price of the synth that inspired it. Sure, the use of digital LFOs and contour generators, and DCOs rather than VCOs, will discourage some purists, but I have great affection for Roland’s DCO-based synths such as the Juno 60, the JX3P and the Super JX10, as well more obscure instruments including the Kawai SX240 and the Bit One, and I would never turn my back on them for this.

So, will Behringer sell a few? As other manufacturers have discovered in recent years, the sub-£1000sub-$1000 market can be a fickle one, but I suspect that if it proves to be reliable and well supported (and I have to thank the Music Group’s UK Innovation team for their excellent support during the course of this review) the DeepMind 12 is going to be a great success. Even before it hit the streets Uli Behringer stated that, due to the overwhelming response, he had already decided to establish a synthesizer division within Behringer, and was hiring engineers and product managers for this. I just hope that the company’s next releases are a little less obviously based upon classics from elsewhere. Sure, we don’t pillory boutique manufacturers for their emulations of vintage synths and modules so, provided that all the relevant patents and copyrights have lapsed, it would probably be hypocritical to take Behringer to task for its use of the Juno 106 as a starting point for the DeepMind 12. But it’s clear that the company have some novel ideas of their own and are capable of implementing them. The future could be interesting.

The Editors

There are three ways to get MIDI in and out of the DeepMind 12: via conventional five-pin DIN sockets, USB 2, and Wi-Fi. This makes it possible to hook the synth up to computers, tablets or even phones, and Behringer are going to take advantage of this with the release of editors for Windows, OS X, Android and iOS. I was sent a link to download a beta version for my MacBook Pro, and after one false start I installed and opened this with no trouble at all. I was then presented with seven main screens — patch overview, preset manager (librarian), preset blender (morpher), effects, control sequencer, arpeggiator, and editor settings — and these not only simplified everything but also added a handful of facilities not present on the synth itself.

The Patch Blender is a four-way patch morpher.The Patch Blender is a four-way patch morpher.

Users are going to like how each sound is laid out in front them and how, when you click on a synthesis section, a larger and clearer version appears for editing. Likewise, having all the effects laid out in front of you is a boon because their parameters make more sense when presented in this fashion, even if this does demonstrate how closely some of the effects are modelled on other manufacturers’ products. The arpeggiator and control sequencer also become clearer when represented in this way, and I can see the four-way patch morpher (called the blender) becoming a firm favourite, especially if this is included in the augmented reality editor. Imagine morphing sounds by moving a marker in mid air using your hands... no clumsy mouse, no pointer, no touchpad... it suggests all manner of possibilities.

The control sequencer in the software editor.The control sequencer in the software editor.

A DeepMind 12 Module In The Wings?

Shortly after unveiling the keyboard, Uli Behringer released a design for a rackmount/tabletop version of the DeepMind 12. He said, “Just to be clear, we have not yet decided to progress with this design,” and asked people for their feedback. I have no idea what was said directly to the company, but the chatter on the interwebsphere is almost uniformly positive. I suspect that, if it appears in this form, it will be popular, especially if supplied (as implied) with both the wooden end cheeks and the rackmount ears. However, the rear panel might need a rethink. The renderings suggest that you would have to leave 1U of clear space above it to accommodate the cables when rackmounted.

I happen to like my analogue synths to have keyboards and controllers because they then become instruments rather than sound sources but, given the precedent set by the likes of the Prophet 6 and OB-6 modules, I would be surprised if something of this ilk didn’t appear. I wonder whether it will have a MIDI overflow facility? It would be a bonus to be able to link multiple modules to create instruments with 24 or even 36 voices.

Augmented Reality

Even before the DeepMind 12 has arrived, Behringer have announced a remarkable extension: an augmented reality interface. If and when released, this will allow you to delve into the synth’s capabilities using Microsoft’s HoloLens, which, in that company’s words, “is the first fully self-contained, holographic computer, enabling you to interact with high-definition holograms in your world”.

The idea is that the HoloLens projects 3D images on to a visor such that they’re overlaid on your environment. If you can imagine these images appearing above the control surface of the DeepMind 12, they could offer possibilities that are not available via the physical controls, the menus, or the real-time controllers. If that all sounds a bit speculative, you should be aware that people who have tried the prototype have described it as “visually beautiful”, “mind-blowing”, and “revolutionary”. I’m particularly impressed with the representation of the modulation matrix, which looks like a huge, futuristic modular patchbay. Being able to connect this thingy to that wotsit by sweeping your hand in space must be a joy but, for the moment, the price of the HoloLens is likely to be prohibitive. Nevertheless, it’s fascinating that Behringer are thinking along these lines. At first I wondered whether anyone other than Daft Punk and their fans would want to go on stage wearing visors and waving their hands in what looks like (and actually is) empty space, but now that I think about it... yes, this could be a glimpse of the future.

Published December 2016