AI & Automated Mastering: What to Know
How does AI fit into mastering? Learn about the different types of mastering available today and how AI and machine learning in Ozone can be used in mastering while keeping you in control.
From ChatGPT and Lensa to folks urging a halt to AI research, we’re hearing a lot about artificial intelligence in our world today. However, at iZotope, we’ve been using AI and machine learning—or ML—since the 2016 release of Neutron. So how does AI fit into music, and more specifically mastering? I’m glad you asked! Let’s dig into it.
Follow along with this article using iZotope
What is automated mastering?
Automated mastering attempts to use computer algorithms to quickly and automatically complete some of the common steps frequently employed in audio mastering. This could include things like setting final levels, setting EQ to improve tonal balance, and even setting parameters for things like compression or saturation. However, none of this necessarily requires artificial intelligence. Rather, it can be guided by the human intelligence of the designer who writes the code.
Thus, while AI mastering is a type of automated mastering, not all automated mastering is necessarily AI mastering.
The success of an automated mastering algorithm depends heavily on the thought put into it by the developer. It’s easy enough to measure the integrated loudness or extract a spectral analysis to determine overall tonal balance, but what you do with that information makes a big difference to the outcome. Generally, automated mastering attempts to apply processing that will match the measured values to preselected targets. Selecting appropriate targets is where it gets tricky.
Some automated mastering systems, like Aria or CloudBounce, require input from the end user to select these targets. To some extent, this assumes that you know what you want and need out of mastering, however, it doesn’t give you that much control over what’s going on in the background. Other systems use AI to set targets, and in fact, this is one of the common uses of AI in mastering. Machine learning makes doing something like genre detection much easier than it would be with a traditional algorithmic approach.
However, one thing that is important to remember is that mastering is about more than just audio processing. With that in mind, let’s take a look at the different types of mastering available and discuss their unique strengths and merits.
What types of audio mastering are available?
There are basically three options: analog, digital, or hybrid mastering—a mixture of both. There are certainly plenty of mastering engineers who work all in the box—or using plug-ins exclusively—these days, and some may even work entirely in analog—using tape as a source, processing through analog hardware, and finally transferring to a vinyl pre-master. Far more common, though, is a hybrid approach in which a combination of plug-ins and analog hardware is used.
In the world of automated mastering systems, hybrid or analog approaches are much rarer. In fact, the only service I know of that uses analog hardware is Aria. All the others use a DSP—or digital signal processing—based approach.
To my way of thinking, I've categorized these mastering techniques by asking two questions: who—or what—is doing the mastering, and what types of tools are being used to do it?
The answer to the first question may feel a bit obvious, but in case it’s not, we’re talking about human mastering engineers and automated mastering algorithms. Honestly, each has its own strengths and benefits—and that’s coming from a working mastering engineer. That might surprise you, but frankly, for an emerging artist, it doesn’t always make sense to pay a mastering engineer. Whether you’re a songwriter shopping a demo around, or a DJ who needs a quick version to play out at a gig, automated mastering can provide a quick and affordable solution.
However, there are a few things automated mastering systems aren’t great at, and that’s part of where human mastering engineers come in. First of all, as mentioned above, mastering is more than just audio processing. A big part of it is the final QC—or quality check—to ensure there’s no unintentional distortion, clicks, pops, crackles, etc. Of course, sometimes those types of things can be intentional, and human intuition is still the best way to judge that.
A few other things human mastering engineers tend to accel at compared to automated mastering systems: track sequencing, spacing, timing, and tops and tails; preparing identical masters of alternate versions such as instrumental or TV mixes; personal communication when they notice something odd in your mix; making meaningful revisions based on your input; and of course, using their experience, intuition, and empathy to bring the best out of your song.
Is online mastering any good?
First and foremost it depends on what you’re after. If you’re shopping around a demo, or need a quick version to try out in your DJ set, sure it could serve a purpose. If you’re preparing something for release, however, I would suggest there are other, better options, including analog, digital, and hybrid mastering options that we discussed earlier.
Online mastering may not be able to capture the nuances and unique characteristics of the music as well as a skilled human engineer—or even a DIY music producer who is looking to master their own music while preserving creative controls.
Is AI mastering any good?
To even begin to answer this, we have to define what we mean by AI. You might think this would be fairly straightforward, but as with most things in life, there are at least a few interpretations. In the broadest definition of AI, it is simply a machine process that simulates human intelligence. In that sense, a sufficiently complex algorithm written entirely by humans—in other words, without any machine learning or neural networks—could be classified as AI.
These days, I would argue that when you say, “AI,” most people think of those latter technologies: machine learning, deep learning, neural networks, and the like. In the world of automated mastering, about one-third of the existing services claim to use AI, but because each service uses proprietary techniques there’s really no way to tell exactly what they’re doing. Suffice it to say, AI mastering is only as good as its algorithms, and not all algorithms are created equal.
At iZotope, we tend to think of this in terms of assistive audio technology. In a nutshell, we use machine learning to help analyze and classify your audio, which in turn allows us to create custom presets for everything from EQ and compression to stereo imaging and microdynamics. This leads us to…
AI mastering with a human touch: iZotope Ozone
At iZotope, we believe that making music is an artistic, human endeavor, and as such our use of AI and ML is in service of that. Our assistive technologies, like Master Assistant in Ozone, leverage AI to do a lot of the grunt work, thereby speeding up your workflow and allowing you to stay in the zone and focus on the details that matter most to you. What’s more, you always retain full control over every element of your master.
Is Ozone better than online mastering?
In my honest opinion: yes. You might think, “Ian, you’re writing for iZotope, of course you would say that.” However, truthfully, I have been saying that since Ozone 8 came out, long before I started writing here, so hear me out. These are my top three reasons why Ozone makes more sense than online automated mastering.
- Ozone does the grunt work for you, but then allows you to get under the hood to tweak any settings you may want. You can even do this in multiple stages. Genre selection and macro controls in the Master Assistant view allow you to dial in the sound of your master from a high level, yet with much more flexibility than any online automated mastering service that I know of. However, if you want, you can also dive into Module View where you can tweak every last setting in every single module.
- Recalls, revisions, alternate versions, and more all become many times easier than with online services. Need to master an instrumental version of a song? Just copy Ozone from the main mix and use the same settings.
- Master Assistant can be used as a learning tool. By inspecting the modules and settings used, and additionally being able to solo or bypass each module, you can start to learn what types of processing benefit your music and in what way. You also have good starting points to work from, so you can experiment and see how subtle changes to things like EQ or compression affect the sound, with the knowledge that you’re already starting in the ballpark.
Start using AI in mastering
With Ozone, you can start using AI in mastering in an open and transparent way that still gives you full agency and control over your music. Automated mastering can be a great option when you just need a quick version to test out, and Ozone’s Master Assistant gives you that in spades. When you need to go further and create the final master for streaming, DJ delivery, or other formats, Ozone allows you to pull back the curtain and get fully hands-on. You can even download a free demo to try this out for yourself right now.
Remember, music is meant to be a form of artistic expression that we enjoy, or at the very least get some catharsis out of. Leaning on machines for a little help so that the spark of inspiration doesn’t get lost in the technical weeds is one thing, giving them the keys to the castle is quite another. Good luck, and happy mastering!