AI Mastering - inevitable or incidental

[JeremyFife]

Came across this article on a brief comparison of 4 "AI" mastering services.

We tested 4 of the most popular online e-mastering services and compared them to a real mastering engineer. Here’s what happened

Are automated, AI-powered mastering platforms actually worth using and how do they stack up against the real thing?

www.musicradar.com

Is this the way that mastering will go? I'm hoping that mastering will remain part of the creative process, with a creative person at the helm.

Perhaps this is inevitable though, like EQ to a target curve?

I'm trying not to generalise, and be rude, and say "it's fine for Pop" ...

One concerning note is they all seem to handle loudness badly, although there's not much detail

 

[restorer-john]

I listened to them all. It's mostly just boom-tizz EQ, a bit of attack compression with some more rolled-off at the top, some more tubby in the mid bass.

Nothing that cannot be done by anyone with Audacity for free.

 

[dfuller]

They honestly all kind of suck, so I'm not worried too much about my job yet.

 

[Blumlein 88]

AI can now master your music—and it does shockingly well

Suddenly, everyone can master their own music.

arstechnica.com

Another article on the topic out today. I tried a couple of these when they first were out, and they were lousy. The samples presented with this article make me think some have improved. The Ozone from iZotope seemed the best to me with this one sample. I might even use it if it pans outs with some recordings I have made. It and the Apple offering seemed not bad. I still think Landr is the worst.

So as mentioned in this article you can have AI write lyrics, come up with band pictures (probably videos if not now soon), write the music, and then master the results. So that seems to leave the human out of the loop altogether potentially other than using taste in judging what the AI has created and adjusting its parameters. Then again a lot of these middle of the road pop female artists might as well be robots considering what they are producing. Now humans can always produce this as a craft/hobby/pastime whatever. That would seem like making music part of craft fairs where people still make by hand various items that are not very cheap or advanced vs factory produced goods. The people like it, and their customers like the human connection.

 

[Cbdb2]

Remember this is mastering not mixing. Usually adjusting the final stereo mix. So a little EQ and compression to the entire track. You cant turn the bass guitar down you have to eq the whole track down from 30 to 2khz. That means kick drum low toms even the lower piano and male vocals also come down. The singers to loud? Good luck.
Same goes for dynamics, the bass is too dynamic (the E string notes are too loud compared to the rest of the bass track) you have to compress the kick drum too.
Things you can do in the mix but not in mastering.
Adjust eq, compression, level, reverb, distortion, panning on individual instruments, including individual drums. More cow bell is possible, not so when mastering.
Doubleing vocals, guitars.
Getting rid of the piano in the intro.
You can even rerecord a part if you find it dosnt work with this mix, or add a part.(this mix could use a harmonica in the intro where we took out the piano)
Removing the first chorus.
Fixing out of tune vocals or a bass thats lagging.
Changing the end fade out to a full stop.
Im sure I missed a few.
So mastering helps but it won't turn a bad mix into a good one.

 

[Blumlein 88]

Remember this is mastering not mixing. Usually adjusting the final stereo mix. So a little EQ and compression to the entire track. You cant turn the bass guitar down you have to eq the whole track down from 30 to 2khz. That means kick drum low toms even the lower piano and male vocals also come down. The singers to loud? Good luck.
Same goes for dynamics, the bass is too dynamic (the E string notes are too loud compared to the rest of the bass track) you have to compress the kick drum too.
Things you can do in the mix but not in mastering.
Adjust eq, compression, level, reverb, distortion, panning on individual instruments, including individual drums. More cow bell is possible, not so when mastering.
Doubleing vocals, guitars.
Getting rid of the piano in the intro.
You can even rerecord a part if you find it dosnt work with this mix, or add a part.(this mix could use a harmonica in the intro where we took out the piano)
Removing the first chorus.
Fixing out of tune vocals or a bass thats lagging.
Changing the end fade out to a full stop.
Im sure I missed a few.
So mastering helps but it won't turn a bad mix into a good one.

Some of these will do STEMs. So yes you have to mix each STEM into stereo (that is what a STEM is), but you can have vocals, rhythm, etc. separate. Same as if you were sending your mix to a human Mastering person. They will do STEM mastering so you do have fairly good control. So no not limited to the entire fully mixed track at once. In this particular sense no difference sending a mix to human Mastering vs AI.

The purpose of AI mastering is not to turn a bad mix into a good one. You cannot do that. It is to create a good master from a mix. You still need a good mix to work with.

Stems and Multitracks: What’s the Difference?

New to audio production? Learn the difference between stems and multitracks, when to use multitracks or stems, and discover tips for creating stems yourself.

www.izotope.com

 

[DVDdoug]

Normally, mixing & mastering requires human decisions and judgment. But AI should be capable of mastering, or it should be soon. If I was using it, I'd want to give the parameters I want, or I'd want the ability to feed it some sample-reference tracks.

It's sad, but AI may replace the artists and everybody else. "Make me song that's derived from the top 100 selling singles of last year", or a blend of the top 100 female pop artists, etc.

 

[dfuller]

I'm still not expecting it to work well, to be entirely honest. Mastering is much more a quality control role than "make it loud" - my most commonly used tool is Izotope RX, not any fancy outboard.

 

[Cbdb2]

Some of these will do STEMs. So yes you have to mix each STEM into stereo (that is what a STEM is), but you can have vocals, rhythm, etc. separate. Same as if you were sending your mix to a human Mastering person. They will do STEM mastering so you do have fairly good control. So no not limited to the entire fully mixed track at once. In this particular sense no difference sending a mix to human Mastering vs AI.

The purpose of AI mastering is not to turn a bad mix into a good one. You cannot do that. It is to create a good master from a mix. You still need a good mix to work with.

Stems and Multitracks: What’s the Difference?

New to audio production? Learn the difference between stems and multitracks, when to use multitracks or stems, and discover tips for creating stems yourself.

www.izotope.com

When I was involved in mixing tv/movies they were all mixed to stems. Dialogue (so foreign delivers could remove the English ), Music, SFX, background fx, foley. We did this to make changes easier. Producers would listen to the mix after it was done and change things. So to change sfx you would only need to match the 16 sfx tracks (instead of all 60 tracks) and there effects before punching in the change on the sfx stem. (all we had was fader automation at the time). Latter on we started to receive music stems, because the composer didn't know what kind of SFX the music would compete with, so taking out percusion during a a scene with percussive SFX could clear things up. A good composer doesn't put a ochestra hit over an explosion or thick heavy metal over whispering. Yould be surprised how often this happens, and how hard it is to deal with.

I havent done this for a while so Im curious, what percentage of music mastering is from stems? It used to be zero. If your working with stems your mixing so this process should be called mix/mastering. Most producer don't want the mastering to change the balance of the Instruments, thats why these where two different procceses.

 

[Cbdb2]

I'm still not expecting it to work well, to be entirely honest. Mastering is much more a quality control role than "make it loud" - my most commonly used tool is Izotope RX, not any fancy outboard.

Mastering is icing on the cake. Bad mix, often from bad recording, or even bad orchestration (too many instruments playing different things in the same freq. band can sound like mud right off the floor) means bad master.

 

[radix]

I'm not up on the generative AI literature for music, but I suspect the main limiting factor is the model of what is an ear-worm or a catchy hook. The better the models of how a person (or subgroup) responds to music, the better generative AI can create it. Once those adversarial models are good enough, AI could begin to generate good novel phrases and patterns that are attractive to some population.

 

[FrantzM]

Hi

It may sound shrill or alarmist but .. when AI would have replaced every profession, craft or human activities .. what will we do?What will be our "raison d'etre"? Seriously? When will AI consider us redundant, useless? Am I one of the few to be very concerned about this AI development?

Interesting that, perhaps, it wouldn't be disease, wars, or even cosmic events that would erase us from existence, but our own hubris....
Sorry , do not have much to add to the thread... Just voicing my fears.

Peace

 

[Blumlein 88]

I would say stem mastering is not the norm. Maybe 5% of the time. But if AI is doing it I'd think it might produce better results from stems vs stereo.

 

[dfuller]

I havent done this for a while so Im curious, what percentage of music mastering is from stems?

Definitely not the norm, in my experience - stems are rare vs receiving a mix. It is genre dependent though, some are more stem focused.