The term “AI” is everywhere these days. It’s in your browser, on your phone, and probably even in your refrigerator. What started as a science-fiction concept in the mid-1950s is now a reality. Kind of. Physical AI that actually aids us in our daily lives with tasks like folding clothes or cooking are at least a decade away.
What we do have is benign, assistive-AI in the way of GPTs and generative tools for music, art, and writing, among other things.
Unfortunately, the vast majority of products and services tagged with the term “AI” are nothing but gimmicks. Most of the time these are just rebranded items that perhaps have some rudimentary sensing technology, at best; there’s nothing intelligent about them – do you need AI in your toothbrush?!?
The marketing schemes don’t stop at consumer goods. “AI” has infiltrated the music software industry as well. For example, it seems that every mastering suite is “AI-powered,” whatever that means.
Enter DataMind Audio’s Combobulator, the “AI neural synthesis” audio plugin.
Is Combobulator a unique sound-shaper based on this emerging technology? Or is it just another audio app using the term “AI” to push more units? Let’s find out in our Combobulator review…
What Combobulator Isn’t
Combobulator will not make music for you
Combobulator is not a generative-AI type application like Google’s MusicFX or Udio. It will not help you write the next Country song about a man’s love for old-school hip-hop. It will not turn you into the next EDM superstar (automatically, at least.)
It is not an AI assistant that will help you rapidly iterate through arrangements and MIDI melodies, or perform menial tasks such as cropping a waveform to a perfect loop, as WavTool does.Check ANR’s review of that software, here.
(WavTool: An AI-Assisted DAW)
Datamind Audio is clear about Combobulator’s capabilities (from the manual):
NOTICE: Combobulator will not make music for you. The AI models are not designed to faithfully replicate the artists they were trained on, or to make you magically sound like the professional music producer your model is named after. Instead, think of it as a synthesis device that has essentially learned how to hallucinate complex timbres based on patterns perceived in the training material.”
What Combobulator Is
Combobulator is, at its core, an AI-powered resynthesis engine that analyzes your input and applies different tones and timbres to it, making it sound completely different, yet retaining the source material’s “feel.” Some have called this a “style-transfer” or “timbre-transfer” effect, which it is, but it’s much more.
There are other, similar products out there, such as Neutone Morpho, but it seems limited to producing sounds from one type of “model,” as opposed to Combobulator, whose “brains” are capable of spitting out potentially thousands of sounds.
Here’s how it works:
Combo (for short) takes an input and analyzes its sonic properties; its pitch and amplitude, for instance. Some conditioning of the signal is done (which can be controlled with the Attack and Sustain controls), then it goes through the Combo engine for processing, and finally to the output.
INPUT (EQ, Shaper)>>ENCODE>>PROCESS (Bend, Scale, Offset, Width)>>DECODE>>OUTPUT
Sounds like a simple signal-flow, but there’s a lot that happens at each stage, and every control has a drastic effect on the resulting sound.
(The Combobulator Interface, as of Version 1.0)
Starting at the top, the Input control is the first stage of pre-processing. You’d be surprised at how much this one control does to the sound. At maximum, any signal will excite the Combo engine and the output will be extremely lively and animated. Near zero, the latents, that is, the neural “strands” have less to act on, so the effect will not be as great. At 0%, you’d expect nothing to happen, but the opposite is true: with no information to act upon, the engine has to make decisions on its own and it all goes a bit “HAL 9000,” but without the homicidal tendencies. More on this, later.
Attack and Sustain are important sonic-shapers. High Attack settings will make the source material more punchy and, in turn, make more percussive selections from “the brain.”
The EQ section enhances certain parts of the program material’s frequency spectrum and, again, this will affect the timbres applied to your track from the brain. It all depends on which brain you’re using, but I discovered a bunch of new sounds when manipulating just these controls.
(Artist Brains, like the ones from ill-esha, are essential add-ons.)
Maybe now is a good time to talk about these so-called artist “brains.” A brain is the essential component of the Combobulator plugin. It’s the data model from which all sounds, tones, and timbres are selected and applied to your sound. As of writing, there are ten available for $33 each, at the DataMind Audio product page: https://datamindaudio.ai/shop/ DataMind has been nice enough to include two of them for free, included with purchase of the main plugin. I’ll go into more detail about each brain later, in the demos section.
Moving on to the right of the interface, you have Encode, Freeze, Bend, Dilation, Dry/Wet, Width, and Output. “Encode” and “Freeze” are brand-new features with the latest version. The manual describes what they do, but it’s kind of hard to grasp. In practice, Encode seems to control how much of your signal gets processed. It seems the closer this gets to zero, the less your signal shapes the output. It’s almost like everything gets smoothed out and you hear the natural sound of the brain.
Bend, and the related control, Dilation, are also new features with version 1.0. These are more sound-shaping “hacks,” mainly used when Encode is near zero. I have noticed more variety in sound-texture regardless of where Encode is, however. More research is necessary.
Dry/Wet and Output are self-explanatory and operate as expected, but Width is a bit different. This control extends the sound-stage by diversifying the timbres applied to the left and right channels. At zero, both L and R channels will contain the same content, but as you tilt the control closer to +/- 100, the channels become completely uncorrelated. It’s one of those knobs that are instantly-gratifying to turn and automate.
Speaking of automation, Combobulator has its own modulation system. If you’ve read my previous review on Noise Engineering’s dynamics plugins, you know that I love when developers include this feature. You can choose LFOs, envelope followers, and MIDI inputs. What’s more, the modulator’s parameters themselves can be modulated. Down with boring, static LFOs!
And now, the centerpiece of the plugin…the uh, center of it! Looking like the cockpit of an interstellar spacecraft, this orb acts as a macro control and vectorscope for “latents,” the heart of the Combobulator’s neural-synthesis engine.
Latents
Latents are the characteristic qualities or features of the chosen model that are manipulated when you adjust Scale and Offset. These latents are affected by the program material, input section, and other parameters. Essentially, they are the ingredients that come together to re-synthesize your audio; the “AI” in “AI-synthesis.”
According to the manual, there are 32-64 latents per model, the first four representing the most dominant features. These four are directly user-accessible with the Scale and Offset controls. There are further details about Latents, Scale, and Offset in the manual, but suffice to say, you control the sonic qualities of the model (dynamics, timbre, amplitude, brightness, etc., etc.) with Scale and Offset. That’s where the real fun begins!
I found myself tweaking for hours, discovering innumerable sweet-spots in my quest for digital gold.
In Practice / The Sound of Neural Synthesis
Writing this Combobulator review took some considerable time, not only because I had to wrap my head around what’s going on internally, but also because the output is so cool and interesting. This happens often with a new plugin or piece of gear. Everything is initially so fresh and new that, before you know it, you’ve recorded hours of material and have 10 projects named “Combobulator test3_1 copy2.” I found myself tweaking for hours, discovering innumerable sweet-spots in my quest for digital gold.
Without a doubt, Combobulator is built for percussive material. Going further, I think you’ll get the most out of it if your source material has some space, atmosphere, and dynamics. The more varied information you feed Combobulator, the more it can act on, and the more diverse the output. For instance, some trailing reverb on the end of a snare hit can yield unexpected perturbations in the model.
That said, with the right combination of artist brain and certain settings, your plain, tonal piece gains new levels of complexity. I started to really enjoy the cacophony of sounds layered upon whatever I threw at Combobulator.
The output is high quality, too. It doesn’t have that smeary, low-bitrate MP3 sound that most AI-generated audio content has. Since the sample-fodder contained in the brains is full-spectrum, high-quality stuff, any audio processed by Combobulator will be, too.
Check out the audio and videos I produced, in the next section.
Now, the way I use Combobulator may be different than other artists. I use it exclusively as an audio effect for my sound-design-heavy music. Realtime performance isn’t important to me, so I was not bothered by the intrinsic latency. Some have reported some heavy delay, but, according to DataMind, it’s an essential trade-off between performance and quality. Combobulator requires a huge buffer to be able to ingest and process your audio data.
UPDATE: This Combobulator review was written prior to the release of “low-latency” versions of brains and a “delay compensation” toggle. DataMind says that these lower-resolution modules can bring the latency to below 100ms. I have been able to verify that, in fact, there is a vast improvement. It’s not quite real-time, but it’s pretty good, all things considered. A bonus is that these new brains bring a whole set of different sounds for you to use.
Demos and Artist Brain Review
Watch this first video I produced, demonstrating the currently-available artist brains. The audio source is Forever89’s Visco sample-modeling drum machine. The video leads with the dry signal, which is sort of a silly, simple beat with some global animation of the “transform” and “contrast” parameters. Note that the audio is sonically rich with variations in pitch, velocity, and other characteristics. Again, material like this is ideal for Combobulator.
Here’s another video showcasing what Combobulator does with more tonal material, using Noise Engineering’s Loquelic Vereor multi-engine synth:
The stand-out brains for me are Grouch, Rhino Betatron, Woulg, and Rob Clouth. Woulg is my favorite, as it specializes in harsh, industrial textures as well as “disembodied voices, granular synthesis, and sonic mayhem,” according to the website.
Eliderp has more of an upbeat, housey vibe to it. I hear lots of stabs, horns, and chords.
The two ill-esha brains are unique; the “vocals”one is made up of her vocal productions. You can apply this to chromatic, monophonic material and Combobulator will track it fairly well, resulting in eerie, time-stretchy vocal textures. It gets super weird when applied to percussive tracks.
I did not have time to dig into Somtatoast’s offering, as it appeared late into writing this.
It’s really difficult to review the artist brains because there are a million parameters that affect the output. I will say definitively that you’re going to want more than just the two that are included. “Rob Clouth” and “Zee Sincorswim” are galaxies full of diverse sounds, but the more you use them, the more you hear the same timbres. You could think of Artist Brains as sample-packs, in a way. Sure, “808 Kit” is great, but it’s nice to have variety.
More Demos
What experimental software review would be complete without a clip from Richard Devine demonstrating it?
Issues
No doubt, using Combobulator has been a positive experience, but technically being beta software (version 0.9.19, at time of writing), there were some minor, mostly UI issues.
We’ve already written about the built-in latency and how DataMind have improved the responsiveness of the plugin by way of low-latency versions of the brains.
The way modulation works is a little weird. For instance, when you attach a modulator’s output to an input via virtual “patch cables,” the default amount sent to that parameter is 50%. I think it’s more intuitive to have this at 100% at the start. Also, I wasn’t completely sure how to remove a modulator from a target, initially. A double-click worked.
The modulation controls kind of float over other UI elements when you hover over them. Maybe a drop-shadow could improve the appearance? I’m not sure. I am happy to see all the active routings pop up when hovering over a modulator, though. It would be really annoying to have all of the “patch cables” visible at all times.
Minor UI concerns aside, the plugin is stable and the good news is that support is quick to reply. (Join their Discord for best service.) Even better news than that, the best news is that they’re constantly improving their software. There’s real passion for this product.
DataMind Audio are on the side of the humans, humans who want to make art for themselves…
Conclusion
DataMind Audio has created a very unique and specific effect. It’s difficult to write an unbiased review on such a thing because everyone will use it differently.
Although I could see it used as an accent for more standard electronic music styles, for artists who want to sound like Richard Devine, Autechre, Venetian Snares, and the like, Combobulator is like sniffing catnip. I mean, it’s the drug that allows you instant entry into a universe of glitchy drums, chopped-up vocal samples, half-heard radio broadcasts, and micro-sound. Put it on almost anything and you add this layer of complexity that would be hard to recreate otherwise.
It begs the question: In a genre where originality and creativity is paramount, such as experimental electronic music, isn’t using something like Combobulator the antithesis of that? I mean, you’re taking your original material and transferring another artist’s flavor and feel to it. Is the idea really “yours” anymore?
Let me put you at ease. The answer is “yes.” The sheer variety of timbres contained in each data-model is vast. Moreover, the sound-shaping parameters further obfuscate the brain’s sample material beyond recognition. What may have been Mr. Bill’s “signature sound” is now a part of a beautiful cacophony, hybridized and intertwined with your material. Conceptually, it’s akin to using samples in your compositions.
And speaking of an artist’s signature sound and samples, I would be remiss if I didn’t mention that each artist who contributed their work for training and subsequent transformation into a “brain” receives 50% royalties for sales of the product. More importantly, 100% of the sounds come from those artists and can be used by you, in any way you see fit; you don’t need to be concerned that the models were trained on copyrighted material, as alleged in some recent cases. The speed of technology often outpaces ethical concerns, so DataMind has done the right thing here.
There’s so much more to talk about regarding AI and how it will change the way we make music, but for now, know that DataMind Audio are on the side of the humans, humans who want to make art for themselves, to be specific.
If my Combobulator review made you curious, you can learn more on the DataMind website.
Reviewer’s Note: This Combobulator review was done primarily with a beta version of the software, so some aspects of the plugin were not able to be reviewed in-depth. This is mainly relegated to the frequently-updated list of artist brains. Go to the company’s website for the current list of products and add-ons.
Pricing
Combobulator’s introductory price is $149. A 7-day trial is available and highly recommended. With that, you get access to a free brain (and its associated low-latency version, if available.) Each additional brain is $33, and again, I’m going to suggest you pick a couple of these up if you buy the plugin.
Written by Derek, a.k.a. (d) / glitched
Derek, a.k.a. (d) / glitched is a multi-genre music producer, focusing on experimental electronic music. You can find his sounds at: http://glitched.bandcamp.com
