What software are you using to extract stems from music?

slotools

What are you using to extract stems?
I have used izotope RX8 since its release, wanted to switch to rip x but my credit cards consider it a fraud account and block purchase.
Anything out there for honorable mention?

peter

I am not satisfied with the quality of any of these apps or services. I don‘t use them and create playbacks myself or buy multitrack playbacks.

SignalProducer

You can buy multitracks from karaoke-version.com (custom backing track section) or jamkazam.com (one song is free after registration). Quality is vary song to song, but catalogues are very big. If you need to extract stems from an audio file - check out moises.ai, separation of instruments is very good and can be an option too.

slotools

Yes stems extracted are poor, I usually replace drums and bass with superior drum 3 and toontrack bass .

Just not happy from what I have heard from karaoke-version, but have not investigated them much.

Thanks for your input.

Ostitu

Moises or Gaudio Studio.

Midimannz

Hello, I use RipX to rip 5 stems at once, then in to my DAW Cakewalk by Bandlab for further editing or transposing.

LeesKeys

I own RipX. It works, but like others have noted, the quality of the resulting audio stems is entirely dependent on the quality of the original recording. In most cases it's not high enough for backing tracks, in my opinion. I've used RipX to create midi stems on a couple of occasions. In most cases, it's a lot simpler to pay $3 for a Karaoke-Version multi-track, load the files in my DAW and tweak as needed.

dmcknight

I've used LaLa.ai with different degrees of success. One was from a demo tape made in 1983 that worked overall amazingly well. Another, I tried to put a new mp3 into it and didn't like the results at all.

HughB

Nothing is more satisfying than doing your own backing tracks from scratch.
You have to stop wanting everything done yesterday and start enjoying music for what it really is.

JohnnySolo

I create my own backing tracks but x-minus.pro is pretty damn good. Very good option for a lazy musician! lol

peter

Yesterday I needed a bit of a distraction from Stage Traxx so I looked at the current state of AI stem extraction.

In 2019 Deezer released an ai model for stem extraction called Spleeter as open source software: https://github.com/deezer/spleeter

This triggered the creation of service like Moises and others that use Spleeter under the hoods. Later followed other services like lalal.ai with custom trained models. But most notably Facebook Research released their own ai model called Demucs (https://github.com/facebookresearch/demucs) in 2021 as open source that immediately won the Sony MDX Challenge 2021 for audio separation.

Since then the development of Spleeter has somehow stalled while Facebook Research improved the model further. They are currently at version 4 of that model.

One of the lead developers of Demucs has left Facebook recently and hired with another company that released another high quality separation model called MDX.

And that is the current state of development as far as I am aware of.

It used to be quite challenging getting those models to run on your computer but luckily there have been some open source tools released that make running these models a breeze. Most notably I can recommend taking a look at Ultimate Vocal Remover: https://ultimatevocalremover.com

This tool makes these ai models usable for everyone. A single click install and you can run most of the open source ai models available on your computer.

So how do they fare?

I got to say that things developed quite a bit since the last time I looked at this topic. If all you want to do is to remove vocals from a track, the UVR-MDX-NET Inst HQ 3 model does an absolutely stunning job. The instrumental playbacks created by this model are nearly artefact free. In my opinion this would already work nicely on stage if all you need is a full instrumental playback. Performance of this model on my M2 MacBook Air would even allow realtime conversion. A 6 minute song needs about 1-2 minutes to process. Unfortunately most models need at least 8GB RAM so running on iOS is not an option not even on an iPad Pro.

Stem extraction on the other hand is a bit mixed. The latest Demucs model htdemucs_ft is way ahead of Spleeter and the best I found so far. It can extract stems for Drums, Bass, Vocals and Other. There is a 6 stem model htdemucs_6s available that also tries to separate Gtr from Keys but that does not work well. Drums, Bass and Vocals extraction is working OK. The stem for Others contains noticeable artifacts. As a practice tool it is working stellar especially for drummers and bass players. It also helps a lot for keyboard and guitar players to be able to hear those tracks more clearly. But using these stems for a modular playback is not yet at a point where I would say it is usable. Processing time on my M2 MacBook Air is near real time but way slower then the MDX model.

So the bottom line: Removing vocals from a track works surprisingly well. Creating multiple stems is still a work in progress. But given a bit more time this also might turn out to work well in maybe 2 or 3 years.

We are truly living in astonishing times seeing all these developments. Download the Ultimate Vocal Remover and play around a bit with the different ai models. It is definitely worth it.