Big shout to Will Russel BSc @kingkoi_official who confirmed much of the LANDR background and operational information below in his dissertation.
LANDR – Apparently Intelligent?
Your vibe matters – and your reference points. There are so many ways to take a finished mix, from precision clarity to slammed analogue clipping. A good mastering engineer will deliver genre leading masters that stand out from the crowd. I studied Artificial Intelligence (AI) for a PhD, it’s at the heart of LANDR, which ‘ machine learns’ what a good master is from a chosen set of good mixes and masters. And it’s good at sitting a mastered mix into that select ‘average’ for the genre! Amazing, … right? Well here’s the issue:
No one will ever learn how to write or edit a great novel from the average word count of best-sellers
AI algorithms are now very good at real world output based on previous data, but rubbish at the things that make art erm … art. Pretty impressive at getting a car to navigate a road, but completely incapable of embracing the summer breeze and flickering sunlight through the trees….
Let’s look at why LANDR and other automated systems might get you decent but ‘average’ sounding masters.
Mastering and mixing is not about flat frequency response and ‘clarity’. Often, quite the opposite. We actually need a hyper reality between those hifi speakers, buds, radio etc. A (stereo) audio field that somehow tries to emulate the buzz at a stadium gig or intimate jazz set And still sounds great thrashing a 120dBSPL club PA system. Think about that … great on a tinny phone and awesome pumped. wtf?
Mastering, like mixing, needs creative vision. Not just technique, kit, or statistical analysis
Great mixes & masters are not achieved through simply ‘hearing all the instruments clearly’. It’s lazy, BS textbook-nonsense that leaves many undergrads in cognitive dissonance. Mixing & mastering is so often unquestioningly assumed as a dull tekkie process in schools and unis – maybe because so few full time staffers have ever earned their living in the creative gig economies. Fact is tech, art, sci & bizz blur in the creative industries but they’re siloed in academia. Agile, fluid and rapid learning is key to freelancer survival. You need your tech chops to deliver your vision, or you fail to get traction. That’s the post-digital, genuinely free-market reality in one of the UK’s (prior to Covid) biggest earning industry sectors. Great mixes and masters need flair as well as highly developed tech skills.
What’s more, we don’t usually like stark digital HD video shot under fluorescent light. We like grain, blurring, and mood. Yes, we prefer creative ‘analogue’ distortion, but only certain types of distortion.
Mix & mastering engineers need to know our harmonic distortion elbows from our enharmonic backsides – One shimmers, one is mostly just unmusical screech
Then there’s the phase distortion of legendary analogue EQ like the beautifully Softube modelled Summit EQF 100 above vs killer high-end linear phase transparent EQ like DMG Equilibrium. When do you use which? If you can’t hear the difference blind in an a/b test, you should not be charging to master or mix tracks. Automated mastering has zero scope for this creative judgement – it doesn’t know anything about what sounds good, or how those great masters and mixes it is trained with were actually made. It just establishes where the frequency & loudness data lie and applies enough control to sit your track in that average. Super smart EQ and loudness match. But anyone that’s ever tried EQ match in Izotope hoping it will magically transform their mix into a blockbuster knows all too well that’s not what happens. A mix or mastering engineer’s ability to hear through a mix – based on experience – is closer to witchcraft than harmonic FFT frequency analysis.
LANDR Is Pretty Damn Smart Though
LANDR is a very smart algorithm that delivers masters with various EQ and loudness options – through adapting Izotope Ozone parameters. Many of us mastering engineers use parts of Ozone. It’s extremely powerful. But LANDR doesn’t understand your artistic intentions. The team behind LANDR at Queen Mary’s have done some important and impressive work, but algorithms couldn’t care less how you want it to sound. That’s why it is remain cheap as a service. I predict that will remain the case – as the best demand the best. LANDR has few to no big credits currently. But it can outperform even some well established mastering engineers; as shown clearly in Will Russel’s paper. What i understood clearly from supervising Will’s work was this:
LANDR will outperform novices easily. It might also outperform the sort of mastering engineers who learnt their trade on great mixes that were trivial to take credit for polishing. But it will not outperform the new breed of mastering engineers who can mix well … as long as they reference properly
Emperor’s New Clothes
That’s impressive LANDR disruption. Some big names in the mastering world may need to up their game. I learned the hard way as an artist how little some (even stellar) mastering engineers care sometimes … and how mediocre the results can be. My big name mastering wizard didn’t deign to reference the tracks I took. Maybe he thought he had ‘golden ears’ or the mega-bucks monitors guaranteed insight. Hubris, most likely.
Fact is, the emperor was bollock naked
The masters sounded pants. I remastered them a year or two later when I’d pulled the wizard’s curtain and figured out how to master as well as I could mix – by referencing objectively. It wasn’t without some crying in my beer. The inter-relationship of mixing and mastering can be mind boggling, Even with an experienced ear and determination, it’s a weird learning curve. At once astoundingly simple and utterly complex.
And LANDR and other automated systems can’t advise you on which huge bass lines are gonna disappear in mono club systems. Or why you might be better off looking at making the guitars fight more with the lead vocal. Or about backing off that lead vox compression attack time, and ramping up the ratio. Maybe swapping that super smooth parallel Fabfilter compression for a 100% wet blueface 1176, with all four buttons in for some grit. They also can’t advise on whether it’s likely to cut well to vinyl, or if it’s the right LUFS to deal with Spotify’s levelling tricks and still bust it out when you drop the 24bit live.
While making up your mind, beware potentially misleading advice and marketing. As an eye-opener, watch head honcho Steve Slate proudly slicing the bottom-end punch off a mix to gain loudness in his hilarious FGX demo – listen to it on some decent full-range monitors or cans. Hear and feel the really low punch disappear with those settings. It’s the critical listening that counts. That’s the head of a good plugin company dropping one. There’s lots of other very bad audio tips out there from those blithely professing expertise.
You can seriously eff-up your masters and mixes when you don’t understand or can’t hear properly due to your monitoring
Another example is the incredible amount of snake oil side-chaining tips out there. Total joke. Side-chaining a compressor over your bass from the kick is not a technique to ‘unmask’ both. It just dips one out. We can distinguish a bass and a kick just fine. They have different time-domain characteristics. – i.e. harmonic envelopes. Simple compression and EQ on the kick and bass are the hallmark of some of the greatest mixes ever made. Maybe save it for side-chaining a mutliband compressor over an 808 boom to dip just the 100 Hz region when the kick hits…
Anyway, back to the point, don’t pay through the nose and don’t expect miracles from cheap or free services like LANDR. Judge for yourself and remember, those who master and mix professionally don’t usually spend their time trying to get ad revenue on YouTube giving away their hard won skills. There really is no magic secret to mastering, .. or mixing. Doing it consistently well takes creativity, honed hearing, reasonably expensive kit & discipline. Mostly it takes time to learn the necessary humility to side-step your compensatory and (over)active hearing.
Many music pros will tell you stories of musicians who can tell you exactly what notes are in a complex chord, and which are slightly sharp – and about EQing stuff on an inactive strip. It’s the fact that our hearing obediently seems to make the changes we’re dialling – just not quite enough! Go easy on yourself, the visual cortex is deeply involved in hearing … and visuals override. It’s why I (try to always) close my eyes for a/b tests.
LANDR is Good, But Kinda ‘Samey’
Yes, you heard it here from good ears, LANDR can sound good – absolutely … but it’s ‘kinda samey’. Its tendency is to make your mix sound like other tracks. And maybe that’s what you need if you’re happy just fitting in with the crowd. Go ahead, it’s cheap and much better than your mate Trev with some cracked VSTs. I’m here to make tracks stand out. A better master, more like you want, every time.
Good Mastering Engineers -Don’t Panic!
LANDR has a lot of solid research behind it, but as I said at the top, all current AI software is hopeless at inspiration, intuition and gut feeling. The life-force stuff that artists need. That crazy stuff bewilders the reductionists & materialists as much as quantum entanglement; it’s just not open to systems-based analysis, game theory, chaos theory , stochastic processes, etc etc… It quite simply does not compute.
In other words, LANDR & other AI automated systems will only consistently blind-beat a good mastering engineer when the mice figure out the stochastic hill-climbing algorithms for the meaning of life. As Thomas Hardy nearly said “Computers won’t self-learn any time soon that sometimes beauty lays not in the things, but in what the things symbolise.”