DataMind Audio launched in 2024 and they’ve already proven to be one of the biggest innovators in the domain of AI audio plugins. Their flagship product, The Combobulator, is a neural audio synthesis VST built for sound designers and experimental electronic music producers.
In this article we’ll share an overview of the interface, including full video tutorials that show you how the program works. Have a listen to some of the high-fidelity sounds the Combobulator can create right within your DAW, with testimonials from artists who contributed to its training.
Near the end, we’ll also share a critique of the greater AI music market, to provide context about why DataMind Audio is so important culturally. Low quality instant-song web apps are spreading like a virus and I believe AI VST designers like DataMind Audio and Neutone will be the antidote.
DataMind Audio: AI models based on Artist Brains
DataMind Audio’s Combobulator plugin is an AI-powered VST that passes audio input through a neural network to synthesize entirely new music textures.
The application’s core AI model has been fine tuned and trained consensually on artists’ music, resulting in what they call Artist Brains. Each brain captures the aesthetic of that artist, providing music producers with a direct path to exploring that same sonic landscape, right within the DAW.
As cofounder Ben Canti explains in the demo reel above, the company’s goal is not to replace musicians. On the contrary, they’ve worked out an ethical AI framework that delivers 50% of all sales revenue to the artists they partner with. DataMind Audio’s Artist Brains are available in an online marketplace, where you’ll find some of the most popular glitch and electronic music producers like Mr. Bill, Rob Clouth, and Woulg.
AI audio wizard CJ Carr of Dadabots / Stable Audio is a research partner and consultant at DataMind Audio. As one of the most influential players in the grassroots AI music scene, his role at the company further underscores the legitimacy of their product.
The Combobulator tutorial: How it works
The Combobulator is available as a VST and AU audio component. This means that in order to get started, you’ll need to own a DAW. If you have questions about the application, join their free Discord channel and ask the team directly. A full user manual can be found here.
Be aware that if you’re a live performer hoping for real-time processing, there’s a roughly 900ms latency between input and output. This is due to the large buffer size required for high-fidelity neural audio synthesis. Combobulator is a better fit for studio production workflows than live beatboxing or timbre transfer on other instruments.
Installing the VST/AU component (Supported OS and DAWs)
The tutorial below explains how to get started in Ableton Live, but according to the site’s FAQs, they’ve tested the audio plugin in every major DAW. I was personally able to load and experiment with the AU component in Logic Pro X without any issue. You’ll be fine running on Windows 10+ or Mac OSX 12+. Pro Tools is the only option that they confirm is unsupported.
The Latent Vectorscope audio visualizer
You’ll immediately notice the centerpiece of the Combobulator. This futuristic audio visualizer has been dubbed the latent vectorscope. It’s your direct view into latent space within the Artist Brain, displaying variables contained within the neural network. DataMind Audio makes a comparison between these latents and watching neurons fire in a human brain.
Latents are connected to audio features within the model, collapsed into single variables. They impact the dynamics, timbre, and more. Each model has between 32-64 latents to work with.
Around the perimeter of the vectorscope you’ll find a circular slider with four concentric rings. Those rings correspond to the first four latents in the Brain. The further you stagger the position of those latents relative to one another, the farther you get from the original input.
The Scale setting controls the relationship between your audio input and the neural network. Turning it all the way down will freeze the influence of your input signal on the latents, while turning it up activates the timbres for maximum variety and novelty.
MIDI input
The MIDI module is a sidechain input, meaning that the signal has to be sent into Combobulator from an external source. It’s important to know that MIDI values are only being used as a trigger. The tonal value of the notes don’t matter, so it’s more like using a drum machine where each note corresponds to a particular hit.
Audio output controls
Music producers might be relieved to find some more familiar components on the interface as well. Gain controls master volume and the Dry / Wet fader layers on reverb. The width knob offsets latents between left and right channels to create a wider stereo image. There’s an envelope with standard attack and release, and an LFO modulator for phase position and waveform amplitude.
Loading Artist Brains with modulation presets
To load an Artist Brain, simply select an AI model from the panel located on the left third of the app. The preset selector is at the top-center and will load custom modulation configs in the right third of the app.
I personally found these presets to be a time saver and a big relief. It lets me experiment with each Brain quickly, using parameters that are pre-approved by people who understand the system better than me. Then I can tweak and make changes from a solid starting point.
Listen to the Combobulator: Bass Music Demo
The Combobulator demo shown above is based on their Eliderp model. Have a listen and watch what happens to the input signal as they stagger the latents with an LFO modulator. The implications are huge for electronic music artists who already have sample libraries and want to push their aesthetic into new territory. I found the Vectorscope mesmerizing to watch but also useful in understanding what’s happening under the hood, in a non-technical visual format.
Ethical AI: Green energy and Fair artist contracts
DataMind Audio is at the top of their class when it comes to ethical AI standards. If you’re concerned about climate impact, their computer runs on solar power as the primary energy source. They use a custom liquid-cooling system to reduce the energy footprint. All of the cloud CPU used for machine learning produce zero-emissions.
Their company has established one of the most generous revenue models we’ve seen. As mentioned previously, fifty percent of sales for every artist brain synth goes straight to the sound designer who contributed to that fine-tuned version of the model.
We’re still in the wild west of artificial intelligence, where companies can get away with training AI models privately without government supervision. For this reason, it’s on each business to decide what approach to take when procuring music for their datasets.
Pushback against low-quality AI song generators
Until recently, the AI music tech space has been disproportionately occupied by web apps. Dozens of companies rolled out instant-song generators and marketed to video content creators, instead of catering to the needs of musicians.
Seduced by the promise of a massive royalty-free music market, startups and their investors hoped they could capture a piece of the pie with AI audio at scale. In that rush to be first to market, all of these companies sacrificed music quality in favor of speed. Maybe they planned on improving the music later, once they had made a name for themselves.
A few of these early web apps were bolstered by multimillion dollar investments, when interest rates on loans were still cheap. Now that the investment landscape has cooled, we’re seeing a change in tide. VCs have also had a moment to look at the numbers and realize this royalty free music play isn’t really working.
The strange thing is that these early arrivals are still being promoted by AI hype influencers in “best AI music generator” lists across the internet. That’s true despite the fact that they’re failing to meet a basic quality threshold for the content creators they’re trying to sell to.
For example, I still see companies like Amper Music mentioned in these types of lists, despite having been acquired by Shutterstock more than three years ago, in November 2020. The company doesn’t even exist anymore!
The blind regurgitation of “best tools” by spam sites and social influencers is a symptom of AI copywriting and black hat SEO writers who want to capitalize on the topic without doing even one minute of research.
Meanwhile, new products and ai models are getting better all the time. Consumers are tired of wasting their time with pseudo-AI generators and have expressed their frustration online.
It’s precisely this glut of low-quality programs in the marketplace that gives DataMind Audio an upper hand, by virtue of their high-quality product and relevance for an actual consumer market. The music production community is keen to experience the power of neural audio synthesis, from a company that prioritizes ethical revenue share contracts with artists.
Visit https://datamindaudio.ai to pick up a copy of the app. Start with a free 7 day trial, after which you have the option to purchase it for $129 (during their launch sale).
This article was not sponsored and there are no affiliate links on this page.