"AI" mixing experiment

  • Thread starter Thread starter K-dub
  • Start date Start date
K

K-dub

Well-known member
I threw caution to the wind and decided to fool around w/ new tools. Specifically Izotope Neutron and Ozone. They promote their AI capabilities. I was skeptical. I applied it in some cases to individual tracks - bass/snare/kick/acoustic/electric guitar etc ... as well as busses drums/keys otherwise.

I followed the results generated by the AI and only minorly adjusted to fit the mix. I didn't make a ton of adjustments because I was interested is determining whether AI was settling on good decisions. I still had level setting and panning to determine. I still had to tweak to suit. But it wasn't a lot. The AI was good about getting me quickly into a neighborhood I could work with.

What do you think of how it came out?
 

Attachments

They are quite different from the other mixes you've done. The 1st and 3rd seemed somewhat bright.

All of them have a different feel, everything is very immediate sounding, very up front. They seem somewhat dry, I don't know of that was your choice or the AI's choice.
 
Yep they're bright, a little bass light.
Hats maybe overly ticky, kick a little clicky

Quite impressed with my rhyming so far :D
 
Agree with you both. That's why I included three examples - so the commonality of mix direction would show.

They are way more compressed and forward than I would have done. Curiously, the mastering plug (also AI) wanted them even hotter. I probably rolled them back from 85% to 60% already (from a setting of 19 to 15) and they're still hot.

I think AI might be trying to deafen us. :D

Overall, though? It's consistent. It's not completely to my liking, but it's not completely to my disliking either. I'm going to work on the same songs this morning and see if I can't get something closer to what I would choose on my own.

These things work by formula - and that CAN help in some ways and wander afield in others.
 
That's probably one of the things that AI has been trained to do... everything went through the loudness wars, so it thinks that's the correct sound. Pump up the volume, compress the crap out of it, brighten up everything until it means a predetermined curve. It's surprising that it also didn't pump the bass, put the "smile on the EQ curve.
 
Redone w/ my overrides on the auto-settings. Mostly I pulled back on the compression of the drums - which tends to create a high frequency cymbal wash, and I pulled back on the high end favored by the AI - which Kieran (Bulls) correctly heard as clicky. I agreed.

The compression settings on these masters are about four-five clicks lower than the AI's choice. I think it creates more air around the sound, which I favor. I also readjusted other levels which needed it after changing the AI's choices.

Mind that I tend to view AI as more of a "personalized" plug-in preset, specific to the track being inspected. Like any preset - you have to tailor it to what sounds best, but it does help to put it into the right neighborhood to start.

The difference on these attached mixes are I wandered further away from the initial AI chosen settings. Put simply, I changed the "presets" to suit my ears better.
 

Attachments

Is that all you? Fantastic slide playing.
It's my band from my "Who We Are" project. I'd been doing project studio stuff since the 90s online, and after I sold my biz in 2016, I wanted to take my material live again. So I initiated a search for local musicians who might be interested in playing original music. Two members from my college band were interested and one of them (bass player) had been working w/ the guitar player - who had never played slide before in his life, but once we asked him to learn (because we were doing a Traveling Wilbury's cover live) he became a fanatic about how much he enjoyed it. The guitar isn't a guitar. He's playing a guitar playing a guitar synth - hence the weird string-like decay in the release.

Aside from the 2 mentioned above - who drove about 40 minutes to practice, the rest of us - the lead singer, drummer, sax player and myself, lived literally w/in a mile of each other.

Can't get more local than that!
 
That's probably one of the things that AI has been trained to do... everything went through the loudness wars, so it thinks that's the correct sound. Pump up the volume, compress the crap out of it, brighten up everything until it means a predetermined curve. It's surprising that it also didn't pump the bass, put the "smile on the EQ curve.
Yep. It's the GIGO principle of computing.

EQ is where the models shine, but curiously, although they look rather pretzel like on the original instrument when the AI analyzes it, the aggregate bus, including the two-bus are remarkably level.

The compression to achieve the desired loudness, though? It's notoriously high.
 
Back
Top