Doing the impossible can be done in 30 seconds now

I've been using the latest version of my DAW (kind of like the music version of photoshop), 'Studio One 7'.

Some of the updates are incredible, as they are every single year.

Among many things, one of the updates is a supposed 'AI Stem Separation'. My eyebrow gets raised almost as fast as the word 'AI' hits my eyeballs, so naturally I'm skeptical. People throw it around like confetti, and more than half the time, it's not AI at all, but just... coding?

Stem Separation is basically where you get any song, put it in a program or website, and it'll isolate the individual instruments and split them into their own tracks. So if you want to only hear the vocals of Toto's Africa - this is how you do it.

It's phenomenal technology, but it's actually not new. I've been using a program called Izotope RX9 to do this for quite some time - long before A.I. was even a term being thrown around.

So what does Studio One mean when they include AI? I'm not sure. They mention a 'Z Plane algorithm'. Z plane is some other plugin company so I guess they're just collaborating or something, but to be honest, the outcome is no different to my RX9. so don't market with AI. It's misleading.

That being said, it is incredible. Assuming you know the song Englishman in New York by Sting, here's the vocals that I just isolated with Studio One:

The whole process took seconds - it works faster than real time playback.

andy-mobbs/2024-10-16-andy-mobbs-englishman-in-new-york-vocals?si=712282000b924eeeb565e0fdd0a1c93c&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing soundcloud ~~~

As you can hear, the entire track is just gone. Drums, pizzicato violins, bass, keyboards.

It even managed to capture the vocal harmony. Whether you like it or not, it also managed to capture the reverb. Whatever's going on, it's intelligent.

There's a little bit of noise bleeding through in a couple of moments, which a producer can quickly clean up after the fact without too much trouble.

You can imagine the uses of this, whether you just want to isolate for learning purposes, or remix, completely re-orchestrate on top. You can isolate vocals and then lower the key to make it easier to sing along to, create karaoke files with the vocals hidden, and so much more.

Wondering how it compares with my old RX9 software, which made no claim to A.I technology, I made this comparison:

andy-mobbs/sting-voice?si=aefc9770bc334589a45971accf84e2a3&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing soundcloud ~~~

It does seem the Studio One version is better, but not entirely. Although overall the difference is very slight, it's also kind of significant, Studio One is clearing out the background really clean, and the harmonies are completely picked up in Studio One but a bit tinny, while RX9's is more full.

Meanwhile, the RX9 version keeps in some of that Seal honking noise after the harmonies, far more than studio one which removes it entirely.

This could all simply be a result of RX9's being a bit less heavy-handed, a bit more loose with its processing, or it could be just time passing and technology generally improving - The latest RX is actually RX11 which I don't have. Perhaps they are superior by now?

Either way, neither are perfect, but I mean come on. that's pretty incredible stuff we can do in a few seconds.

I think Studio One has the capability to add a lot more features. Both of these programs only give you four track options: Vocals, Bass, Drums and 'other'.

The 'other' is a bit of a bummer. I'd have LOVED it to be able to use its magical, so-called AI to identify, say, the saxophone, the piano, and separate them. I'm sure it's not too far around the corner in musical production tech, assuming the music industry doesn't get entirely wiped out and replaced by AI generated tracks churned out in Chinese music factories a billion per day, per room.

Regardless, I'm going to have a TON of fun with this - and hit me up if anyone here wants me to isolate any tracks for performance purposes or whatever. It genuinely takes about 30 seconds so it's no hassle at all!

(I can spend a bit more time professionally cleaning it up to perfection but that'll cost cuz it takes way more time!)