Rush Demucs (AI Separated Stems)

[MarkMerkMork]

I've been running some Rush songs through Ultimate Vocal Remover via the htdemucs_6s 6 Stem Model, which gives you stems of Bass, Drums, Guitar, "Other", Piano, Vocals from a stereo WAV file.

To keep everything organized, I've copied all other links below, sorted by Album:

Grace Under Pressure:
Red Lenses

Power Windows:
The Big Money
Grand Designs
Mystic Rhythms (Re-done with UVR, original LALAL.AI version below)

Hold Your Fire:
Open Secret

Presto:
Available Light



Original Post:

Hey all,

So glad I discovered this forum, I love bootlegs, and I love Rush. Best of both worlds! Props to By-Tor X-1 , your archival work here is incredible!! Hope this is okay to post, please move or take down if not.

Recently started playing around with various AI stem separators, so far it seems like LALAL.AI is the best of the bunch that I've tried. It's a paid service, so I figured I would upload my results here for fellow Rush fans to take a listen.

LALAL.AI lets you separate using the following presets:
Vocal and Instrumental / Voice and Noise (these are pretty similar...the noise one would be more suited for video work)
Drums
Bass
Electric guitar
Acoustic guitar
Piano
Synthesizer

You have to choose the preset individually, it then spits out two files: the "split", and the "no split". I separated them all (minus the noise preset), to see how it would react with Mystic Rhythms.

So what's the purpose behind an AI Music Separator Service? DJs, Remixers, and Producers benefit from this mostly. Video editors that have an interview where there's too much background noise. People that like karaoke tracks could also benefit from an AI service like this. For me, I just love hearing multitrack and stem files! I'm a studio and live sound engineer by trade, so I find it fascinating hearing different production techniques. I have a collection of Rush stems from Rock Band, Guitar Hero, and Jammit if this forum is interested, I can upload them.

My other reason besides enjoying a good stem listening session: I'm a drummer of 15 years and love playing Rock Band (have my Roland TD-11 rigged up to it), and would love to see more Rush custom songs charted. You can chart your own songs with either a stereo file or a multitrack session, and figured it would be fun to have proper instrument separation. Don't think this worked well enough for this application, but I'm going to try mixing them in Logic and see if I can blend the Splits/No Split files to create something useable!

As a whole, they aren't amazing...but the drum stem, vocal stem, and instrumental are a pretty cool listen! I have ~200 minutes left on my LALAL.AI account (you pay per minute per preset that is processed unfortunately, not per song...) so if there's any other Rush songs you want to see here let me know!

Link:
https://drive.google.com/file/d/1ui6_LUeZHOPkJnCnyFl3LMxXWtL7_7nO/view?usp=sharing

 

[Dave]

so if there's any other Rush songs you want to see here let me know!

“Red Lenses”

 

[Grace :)]

I've found Ultimate Vocal Remover to be really good. I use the GUI version. It's also completely free.

 

[MarkMerkMork]

I've found Ultimate Vocal Remover to be really good. I use the GUI version. It's also completely free.

In the time I first tried this to now, the machine learning models used to make these stems have got even better. Loving UVR, thanks for the suggestion!

“Red Lenses”

Only 7 months late (sorry forgot about this thread)...but hey, like I said, the training models have only got better! This is using the htdemucs_6s model

Red Lenses:

Red Lenses Demuc.zip

drive.google.com

The Big Money:

The Big Money Demuc.zip

drive.google.com

 

[cynicsage]

Wow this is awesome, thank you for sharing! Could you please do Grand Designs and/or Open Secrets??

 

[Spetts]

Vocal and Instrumental / Voice and Noise (these are pretty similar...the noise one would be more suited for video work)

If it's not too late to make a suggestion, reading this gave me an idea. Could you try using AI to remove the noise from this interview? It would be great to salvage even a few more words from this flawed recording.

As for song suggestions, I would love to hear Available Light, but I worry that there's not much audio signal for the AI to work with. Presto as a whole sounds so treble-heavy and degraded that I genuinely thought my audio setup was broken when I played my CD for the first time. Full disclosure, I'm trying to make a cover of the song, and I can barely make out what the band are playing at points, especially the bass and kick work.

 

[MarkMerkMork]

I've updated the original post with all other links to keep it organized. Also added is a re-done Mystic Rhythms with UVR to match the others. Enjoy!

Wow this is awesome, thank you for sharing! Could you please do Grand Designs and/or Open Secrets??

Open Secrets

Grand Designs

I think at some point, I may go through and do the whole discography. I only run a Mac Mini M1 though so it's not the fastest process. And to do it right, I'd want to fix issues where UVR isn't quite getting it right, like putting synth parts on the "Piano" stem, some vocal BGs I've seen fall on wrong stems, etc. Right now this is just running UVR on htdemucs_6s with the default settings and I upload it. With some proper processing I'm sure it could be better. Big emphasis on may, it would be a huge undertaking to do well. Unfortunately, Time doesn't stand still lol.

If it's not too late to make a suggestion, reading this gave me an idea. Could you try using AI to remove the noise from this interview? It would be great to salvage even a few more words from this flawed recording.

As for song suggestions, I would love to hear Available Light, but I worry that there's not much audio signal for the AI to work with. Presto as a whole sounds so treble-heavy and degraded that I genuinely thought my audio setup was broken when I played my CD for the first time. Full disclosure, I'm trying to make a cover of the song, and I can barely make out what the band are playing at points, especially the bass and kick work.

Unfortunately this interview is just too garbled to work well through this process, but I've included both so you can hear the results. I actually think UVR did a better job than LALALAI. Side note: I think Neil is saying "College Door" and not "Polished Door" as By-Tor X-1 transcribed, would make sense given the context? (I'm very impressed how much he transcribed from what little clarity remains in that file lol). When it gets rid of the noise it does a decent job, especially certain parts where Neil is speaking.

1976-05-30 Interview

Not too bad on Available Light, I think it will be serviceable enough to pick out what they're playing for a cover!

Available Light

 

[Spetts]

Unfortunately this interview is just too garbled to work well through this process

Ah well, thanks for giving it a try.

Not too bad on Available Light, I think it will be serviceable enough to pick out what they're playing for a cover!

Wow, that turned out way better than I expected! The drum track in particular is super insightful, I couldn't make out the kick drum in the original track.
The bass track in particular has moments that make me ask, "where the heck are they pulling this from?" But hey, that's the nature of this technology. Very useful, but always remember to double check with the real deal.
Come to think of it, I found myself having to double-check the final mix even with official stems! I spent hours listening to the 2112 Overture/Temples of Syrinx stems from Jammit before realizing that it was a live version. Even when you do have the exact same studio recording, any given instrument can sound very different in the context of a full mix.

 

[Analog_Orange]

I've been running some Rush songs through Ultimate Vocal Remover via the htdemucs_6s 6 Stem Model, which gives you stems of Bass, Drums, Guitar, "Other", Piano, Vocals from a stereo WAV file.

To keep everything organized, I've copied all other links below, sorted by Album:

Grace Under Pressure:
Red Lenses

Power Windows:
The Big Money
Grand Designs
Mystic Rhythms (Re-done with UVR, original LALAL.AI version below)

Hold Your Fire:
Open Secret

Presto:
Available Light



Original Post:

Here’s my 45 minute attempt of a remix of Open Secrets

Rush Remix

Shared with Dropbox

www.dropbox.com

 

[sttngds9]

Hello, I loved this topic, I have used several applications like this such as Ultimate Vocal Remover, Lalal.ai and MVSEP.
Which would be best? I've already done Hold Your Fire, Grace Under Pressure, Power Windows, Presto and Permanent Waves in full by MVSEP, which was considered second best in the competition https://www.aicrowd.com/challenges/sound-demixing-challenge-2023 and Every month they update the tool and it increasingly removes instruments with greater clarity.

 

[sttngds9]

If it's not too late to make a suggestion, reading this gave me an idea. Could you try using AI to remove the noise from this interview? It would be great to salvage even a few more words from this flawed recording.

As for song suggestions, I would love to hear Available Light, but I worry that there's not much audio signal for the AI to work with. Presto as a whole sounds so treble-heavy and degraded that I genuinely thought my audio setup was broken when I played my CD for the first time. Full disclosure, I'm trying to make a cover of the song, and I can barely make out what the band are playing at points, especially the bass and kick work.

I will try with MVSEP and send here.