Random Gaga Filters (plus requests)

[gagzus 16,709]

Hey all been playing around with filtering lately and just wanted to share some of what I’ve done thus far. If anybody has any requests for any filters or anything from any Gaga songs let me know!

 

[ALGAYDO 33,875]

Read the title as “fillers” at first and thought this was gonna be an thread about her face lol 

[llemesm 1,771]

i thought you mean this:

[hELXIG 44,091]

How the hell do you do this?? I so badly want to be able to separate out the vocal layers

[gagzus 16,709]

2 hours ago, hELXIG said:

How the hell do you do this?? I so badly want to be able to separate out the vocal layers

I do a combination of filtering out vocals by playing with the old school Invert trick, playing around with editing the frequency, trying to edit out the reverb with DeNoise and sometimes in extreme cases (because I have a good ear for hearing vocal layers) I’ll even record myself singing layers I hear and melodyne my own voice to match what I hear exactly and then use an AI module to transform my voice into Gaga’s and use that filter out.

If you listen to them with headphones you’ll hear where it gets a bit choppy and stuff from all the messing around with a combination of all of them

Edited January 8 by gagzus

[Kayi 2,845]

Zombieboy would sounds even greater if she made it a little more dark.

Funky beats are really goth but I wish it was a little more dark.

When you listen to these layers you clearly hear how great her vocals are. It's almost like rediscovering Mayhem again. 

[spookytanooki 1,197]

9 hours ago, gagzus said:

 I’ll even record myself singing layers I hear and melodyne my own voice to match what I hear exactly and then use an AI module to transform my voice into Gaga’s and use that filter out.

As someone who has spent a potentially insane amount of time doing filters over the past 17 years, I dont think this is possible, unless you're using some kind of technique I've never heard of???

Like, you cant even use separate vocal takes from the same artist during the same recording sessions to accurately filter to this degree, because each individual recording creates a unique waveform.

I cant wrap my head around AI vocals coming even close to something you can use to filter with. Can you please explain the process???

[hELXIG 44,091]

On 1/9/2026 at 1:11 AM, gagzus said:

I do a combination of filtering out vocals by playing with the old school Invert trick, playing around with editing the frequency, trying to edit out the reverb with DeNoise and sometimes in extreme cases (because I have a good ear for hearing vocal layers) I’ll even record myself singing layers I hear and melodyne my own voice to match what I hear exactly and then use an AI module to transform my voice into Gaga’s and use that filter out.

If you listen to them with headphones you’ll hear where it gets a bit choppy and stuff from all the messing around with a combination of all of them

Oh I've used AI to separate the layers the best it can (not great tbh, It's very good at separating vocals from instrumentals but not good at separating vocal layers) and done the invert trick with lots of different filters to chop together some layers before but, I could never get anything this clean. But using your own vocals with a Gaga AI model is a genius way of extracting layers. I have an very good Gaga model (actually 3 different ones I trained myself lmao) but I can NOT mimic her well enough on pitch enough to do that  

I'm kinda surprised that even worked since doing the invert trick seems like it has to be extremely accurate. I recently tried for hours to take the studio instrumentals that leaked recently, and invert them against the official tracks, and was shocked to see how badly it worked. I honestly though I'd hit invert, mix and render the tracks, and have a perfectly clean studio acapella but it was badd   then I realised there's something weird about the leaks. Like they didn't quiiiiite match up with the released songs. Not doubting they're the studio leaks though, just.. idk it's hard to explain. I zoomed in as far as I could and it seems like they just don't quite match on a microscopic level

Also I have no idea what playing with the frequencies does because I literally know nothing about music except for a few things I've picked up playing in Audacity with my AI conversions etc the last year lol but I'm curious how that works

 

Edit: maybe the leaked instrumentals we got are the unmastered versions. So maybe the amount of instrumental that leaks through when you do the invert trick are the changes that were made after they were mastered  but that also doesn't explain why all the leaked instrumentals all start about half a second before the released songs   my brain hurts

Edited Monday at 04:13 AM by hELXIG

[zevthepaparazzo 1,285]

One time I managed to filter out some synths from poker face with an invert using the drumless chorus just after the bridge but then I remembered the stems leaked in like 1783 

Edited Monday at 05:06 AM by zevthepaparazzo

[gagzus 16,709]

10 hours ago, hELXIG said:

Oh I've used AI to separate the layers the best it can (not great tbh, It's very good at separating vocals from instrumentals but not good at separating vocal layers) and done the invert trick with lots of different filters to chop together some layers before but, I could never get anything this clean. But using your own vocals with a Gaga AI model is a genius way of extracting layers. I have an very good Gaga model (actually 3 different ones I trained myself lmao) but I can NOT mimic her well enough on pitch enough to do that  

I'm kinda surprised that even worked since doing the invert trick seems like it has to be extremely accurate. I recently tried for hours to take the studio instrumentals that leaked recently, and invert them against the official tracks, and was shocked to see how badly it worked. I honestly though I'd hit invert, mix and render the tracks, and have a perfectly clean studio acapella but it was badd   then I realised there's something weird about the leaks. Like they didn't quiiiiite match up with the released songs. Not doubting they're the studio leaks though, just.. idk it's hard to explain. I zoomed in as far as I could and it seems like they just don't quite match on a microscopic level

Also I have no idea what playing with the frequencies does because I literally know nothing about music except for a few things I've picked up playing in Audacity with my AI conversions etc the last year lol but I'm curious how that works

 

Edit: maybe the leaked instrumentals we got are the unmastered versions. So maybe the amount of instrumental that leaks through when you do the invert trick are the changes that were made after they were mastered  but that also doesn't explain why all the leaked instrumentals all start about half a second before the released songs   my brain hurts

The AI thing is tedious and that’s why I rarely do it, because it requires me to be almost note perfect (sometimes chopping together multiple takes to make it work) and using melodyne but then it has to be close to my voice otherwise it becomes auto tuned as f- 

Im playing around with using the model I have to make full remakes of songs in her more modern voice because her inflection has changed in songs so it requires having to sing it how I imagine she’d sing it now which can be a pain. But that’s my next major project I’m trying to do Hallelujah specifically since I’ve fell in love with that song lately

[hELXIG 44,091]

11 hours ago, gagzus said:

The AI thing is tedious and that’s why I rarely do it, because it requires me to be almost note perfect (sometimes chopping together multiple takes to make it work) and using melodyne but then it has to be close to my voice otherwise it becomes auto tuned as f- 

Im playing around with using the model I have to make full remakes of songs in her more modern voice because her inflection has changed in songs so it requires having to sing it how I imagine she’d sing it now which can be a pain. But that’s my next major project I’m trying to do Hallelujah specifically since I’ve fell in love with that song lately

Well if you're looking to train another model I have all the data you might need. I originally trained one of a bunch of her vocals, then got this obsessive idea in my head that I wanted to convert The Fame era songs to her Mature voice and vice versa.. to I trained one model off her TF era vocals and one off her 'mature' vocals between Joanne, Chromatica and Love For Sale.. I used them to create these..

 

[gagzus 16,709]

14 hours ago, hELXIG said:

Well if you're looking to train another model I have all the data you might need. I originally trained one of a bunch of her vocals, then got this obsessive idea in my head that I wanted to convert The Fame era songs to her Mature voice and vice versa.. to I trained one model off her TF era vocals and one off her 'mature' vocals between Joanne, Chromatica and Love For Sale.. I used them to create these..

 

I love how accurate these are, but the only problem we run into is that she pronounces words and specifically vowels much different now (see how she used to say “love” as lu-wv like in LoveGame now she sings it like like la-uhv in Perfect Illusion) 

This is just a quick take of me singing Hallelujah to how I imagine she’d sing it now obviously this isn’t tuned but still you get the gist

https://vocaroo.com/1ljaNrH7WYVx