Why I’m Not Afraid of “Automatic” Mastering Software Stealing My Job Anytime Soon
Automated mastering programs like LANDR are here to stay. But are they really a threat to mastering engineers?
New technology. It’s big, it’s fast, it’s scary, and it’ll “take your job“.
So why am I—a guy who makes a great portion of his living mastering records for other people—not at all concerned about the existence of LANDR and other “automatic” mastering software like it?
Why have I been able to keep on increasing the amount I bill in mastering work, each and every year, at the exact same moment in history that such technologies have emerged and proliferated? Why am I able to sleep soundly each evening, my dreams uninterrupted by nightmares of automated, audio engineer Terminators eviscerating my livelihood?
Well, there are several reasons. We’ll have to get a little philosophical to answer these questions. But I promise it’ll be fun, and that it may hold some useful lessons for you, too—even if you don’t master recordings for a living. Let’s dive a little deeper.
Reason #1: Technology Has the Power to Create as Many Jobs as it Destroys
First, let’s take the extreme case, one which I don’t believe is true for a variety of reasons that we’ll get to in a minute.
For now, let’s pretend—for just the moment—that auto-mastering software already sounds just as good as a great mastering job does. Let’s pretend that it can fill the exact same niche, and effectively play the multitude of roles that a really good mastering engineer does.
I think I can provide you with some pretty undeniable reasons as to why the above isn’t even remotely true. But even if all that were true, I still wouldn’t be scared. Why? Because if we ever were to get there, every old job destroyed is a potential new one created.
Over one hundred and fifty years ago for instance, more than 90% of Americans worked on farms. They had to just to survive. Eventually, along came the tractor and mass-produced fertilizer, and farm jobs were absolutely decimated. It may have been a hard transition for some, but Americans became more prosperous, not less. They soon had more to eat, and often, they were soon able to take on even more creative and interesting jobs that this new abundance allowed for.
Today, more than 90% of Americans don’t work on farms. Wow. All those farms jobs gone. Yet here we are. Working. Often, at more comfortable tasks that couldn’t even be imagined before the tractor—and that wouldn’t have been economically feasible before that new technology arrived.
The same occurs in music all the time. What happened to all the piano builders and piano tuners? Maybe, some of their modern counterparts are involved in manufacturing digital pianos that practically anyone can afford. Maybe, some help create magnificent-sounding new software libraries that emulate real pianos. Maybe, some have gone on to do something else entirely that suits them even better still. And maybe some of them are still plugging away at Steinway, creating the next generation of high-end acoustic pianos.
Today, musicians have to cope with the reality of virtual guitarists, virtual drummers, virtual pianists (and now, virtual mastering engineers). Does this mean fewer guitarists and pianists and drummers? Maybe. But does it mean fewer musicians? Maybe not.
Someone still has to program all those virtual keyboards and guitars and drums, don’t they? Someone still has to pick which of those virtual mastering algorithms sounds best, tweak them as tastes change, and help artists pick which ones best serve their art, do they not?
Sure, maybe the marginal, mediocre session guitarist who played on demo tracks by emerging artists is now out of work thanks to the reality of robo-guitarists. But maybe he has a new job, helping to program the robot backing tracks for emerging artists’ demos today!
Even if these new, algorithmic “job-stealers” pushed the marginal drummer or mastering engineer into a different job (or just a different aspect of music creation), can that algorithm replace a truly unique and inspiring artist? One with a distinctive voice, who bends the rules, and invents entirely new sounds? Probably not. Which leads us to the next point…
Reason #2: Art Is About Changing The Rules, Not Following The Old Ones.
There are musicians out there who sound distinctive, who have created a voice all their own— one that is instantly recognizable as uniquely human and uniquely theirs. Which ones come to mind for you?
Let’s narrow it down to guitarists for the moment: Vernon Reid. Django Reinhardt. Marc Ribot. Carlos Santana. Wes Montgomery. Brian May. Jimmy Page. Tony Iommi. Even the guitarists of bands like Sonic Youth or My Bloody Valentine. Pick your poison. (Or pick an entirely different instrument or genre.)
Sure, once these artists have developed their voice and put out a library of work with it, you could potentially come up with a computer program that would emulate them surprisingly well!
Such a computer program could look for rules and patterns in their playing and their tone, and then make new solos and even new songs based on those discovered rules. But can you imagine an algorithm or learning machine that could have invented any of these styles in the first place?
And that’s the problem with computers: When we’re talking about algorithms, someone has to program them. And who the hell would have ever programmed an algorithm in advance that might sound like one of these artists? Who could ever program an algorithm that might spit out a flabbergastingly bold artistic decision like Neil Young’s epic one-note guitar solo on “Cinnamon Girl”? What computer could possibly think of something so stupid and make it so beautiful?
The thing about computers is that they’re all about following the rules. Often, better than humans ever could. But the thing about humans is that they’re all about creating new rules.
It’s not even just that we’re all about “breaking” the old rules. (Though we often do that as well, especially in art.) We invent entirely new ones. When the old sandbox is no longer any fun, we might build a new and previously unimaginable one… Or find an entirely new game to play altogether.
Even at their most powerful, today’s most advanced computers are great at playing games that humans have already created. Yes, a machine-learning A.I. program can whip your butt at chess, checkers or the game of “go“. But human beings can create new games that humans like to play with eachother. Computers are wonderful at following the old rules of the old games. Human beings are amazing at inventing new ones.
Take mastering for instance: What’s enough treble? What’s too much? How about bass, or compression or limiting or widening? Or take mixing: How much reverb is enough on your snare drum? How much is too much?
Well, it depends on what year you’re asking me in. Is it 1970 or 1990? 2008 or 2018? What game are we playing? The answer may be very different. And the answer may be even more different still, depending on whether you want to embrace the trend or buck it.
How much will be “right” in 2028? Who knows? But I can tell you one thing: People will almost certainly get tired of what is “right” in art now, and try something new. Computers, for all their strengths, are abysmal at doing things “wrong” and making it turn into the new “right”.
Whatever you do in music, let’s play a little thought experiment: I just created an algorithm that sounds just like you do. It’s extremely popular. Now, everyone’s records can sound a whole lot like yours, and maybe they do.
What does any self-respecting human do next? Well, we throw a wrench into the works and create a new game, naturally. The old style is now cheesy, and will get your ass fired. The new one will win you praise, accolades, all the bling and all the honeys—if it turns out that kind of thing is still even hip anymore. (Is it?)
And that’s thing: Not only can humans adapt to the ever-changing reality of aesthetic preferences more quickly than any algorithm can (provided they have the right mindset); they can also create the conditions that others have to adapt to.
At this point, you may be saying, “OK, I get it, but you don’t have to keep on using so many italics to drive the point home.” Well, that’s an aesthetic choice. Do you think a computer can make it for you? And what do the computers do when tastes change?
Fortunately, these are the kinds of conversations you can have with other human beings while making art together. Which brings us to the next point:
#3: Someone Else’s Algorithm Can’t Give You Truly Personalized Coaching or Feedback, and You Can’t Give it to the Algorithm Either
Human beings crave human interaction. We are social beings.
In a way, I’m interacting with you right now. Give me feedback! Tell me where I’m wrong. Let me know what I did right. Tell me what I can do better. I’ll read it. I’m interested.
I take the same approach into my mastering work as well. Every single master I do emerges from a conversation with the artist. What records do you love the sound of? What does “the right amount of bass and treble” sound like to you? Better like this, or better like this?
My end results also emerge as part of a larger conversation with every other record-maker in the world, and with every other record on the shelf.
Where should this record fit into the great literature of recorded music that’s already out there? What does this one have to say that those other records don’t? Where should it live compared to the others, and which ones does it want to be seen with, or apart from?
Is there any algorithm that can answer these questions? No. But it can sure give you the same exact low-end profile as every other record that goes through that algorithm. I don’t know about you, but that doesn’t sound like a recipe for artistic success to me.
It’s not just the artist or mixer who can give me feedback either. I give them personalized mix feedback too, whenever they want it. Can an auto-mastering plugin do that?
What could you have improved? What are your monitors or your room telling you that are lies? Are you compressing too much—or not enough—relative to your goals for the end result?
What could you have done, specifically, to get that kick drum or vocal to sit more like you wanted it to? What algorithm is going to answer these questions? Given that they are in large part subjective, what algorithm ever could?
But that’s not all. You also can’t give your auto-mastering algorithm personalized feedback either, sending it away to think about and experiment on how to better realize your unique aesthetic vision. Which brings us to our next point….
Reason #4: Mastering Algorithms Just Don’t Sound That Good Yet.
OK, maybe I should have put this one first. And maybe a computer algorithm would have done so for me! But I had a different preference.
I’ve done listening tests with LANDR and similar services. And here’s the thing: They don’t actually sound all that terrible! Unfortunately, they don’t sound all that great either.
I could link to any one of a number of listening shootouts here, but out of sheer loyalty and good taste, I’m going to link to one by a regular SonicScoop contributor, Jason Moss of Behind the Speakers.
When Jason posted a listening test between his level-matched mix, an automated LANDR master, and a professional master from a capable and experienced audio engineer, I leaped at the chance to listen. You can listen to it for yourself here, and draw your own conclusions.
In the meantime, I’ll tell you basically the same thing I told Jason: To my ear, the professional master easily sounded best, followed by his original mix, followed by the LANDR auto-mastering version, which came in last place.
None of them sounded bad by any stretch. (It’s a good mix!) It’s not that the LANDR sounds terrible or anything like that. I would simply put it in third place for my tastes, and say it sounded slightly less good than the original mix—just a bit louder and wider and less interesting.
You do get a little more width out of the LANDR version compared to the original, which is nice. But otherwise, it just feels a little more plastic-y in the attack than the original mix, and a little more cloudy overall. The professional master by Howie Weinberg however, sounds superior to both the other versions to me in 4 key ways:
1) The attack on the drums sounds way better on Howie Weinberg’s master than on either of the others. They just feel more vibrant and alive. This was one of the first things I noticed, and to me, is worth the price of admission alone.
2) The vocal effects sound way better in Howie’s master than on either of the others as well. You really hear the reverb and delay tails come up in a super pleasant and intriguing way. The vocal effects’ tails come up in the LANDR master as well, but they just don’t sound as good as Howie’s. Once again, they were too thick and “plastic” sounding to me, for lack of a better word. In Howie’s master, you get more detail in the effect tails, instead of more “slop”, like in the LANDR version.
3) The overall EQ curve is easily best in Howie’s master for me. It’s just leaner and cleaner while still sounding really full. You get much more separation between elements in Howie’s master, while still getting power and impact. If anything, I feel like you lose some separation in the LANDR master, as well as some of the vibrance and impact.
4) The stereo spread sounds best in Howie’s to me as well. There is a small flaw here though: The left guitar pulls maybe a little too far to the left. But even with that taken into account, the stereo spread here sounds the most interesting and impressively wide to me.
And here’s the wonderful thing: If I were to tell a human being like Howie Weinberg about my feelings on how he could further improve aspect #4, he could easily take the feedback and tweak it just slightly to make it even more perfect. I do this same kind of thing for my own clients whenever they like.
You can’t really do that with an automated mastering algorithm. But even if you could, it would no longer be an automated mastering algorithm, would it?
As soon as you get into having to make choices between which version of the algorithm sounds best, and then perhaps further tweaking these algorithms to improve their results on your track (preferably while in a listening environment that’s especially well-suited to the task) we’re no longer talking about automated mastering at all, are we? We are talking about computer-assisted human mastering which surprise, surprise, is what we already do.
To be fair, even without the ability to adjust it, the LANDR version here doesn’t sound terrible by any stretch. It’s fine. But to me, the original mix sounds better. So what’s the point?
Because I don’t think it’s advisable to spend money to make your mix sound slightly less good than it did originally, if it was my mix, I’d just as well add a light limiter to it and go with that—maybe adding tiny skosch of stereo widening if you like that effect.
In other cases, perhaps the LANDR version would have sounded better than the original mix. Maybe it even sounds better to you now. That’s OK. But how are you going to know unless you take the time to critically listen to each version, on a system that really reveals the differences, knowing just what to listen for?
If you’re going to go through all that trouble and take that much effort, why not learn a little more about mastering so you can better make those decisions? And, if you’re really obsessed with getting it right at this stage in the process, why not work with a mastering specialist to help you make those decisions—or even work to become one yourself, and help make those decisions for others?
And that’s why I’m not afraid of LANDR, or any other auto-mastering program. When they’re automatic, they can’t help you make great artistic decisions. And when they can help you make great artistic decisions, they’re not automatic.
Justin Colletti is a mastering engineer, writer and educator. He edits SonicScoop.
Please note: When you buy products through links on this page, we may earn an affiliate commission.