Hi folks,

I'm desperately looking for a piece of software that can use input data, be it an acapella, to learn from it and then apply phase cancellation to its counterpart, be it the full song of that acapella, using frequency, time and dynamics altogether. The reason I'm asking this is because many times, acapellas from full tracks don't always align perfectly with each other because both don't came from the same mix. Sometimes one will also not be stable in speed, maybe because of wow effect, aswell as having other digital/analog introduced artifacts.

In short, I want to tell such program: "hey, take this acapella, and use it to remove similar frequencies respecting time and dynamics on this other track, then enhance it with, say, AI technology". All existing programs I know aren't able to do this, as they're limited to static noise profiling.

If any ideas, please share your knowledge. Also, please share how would you approach the extraction of a vocal in such scenario: acapella IS the same one as the original track, but both have differences in mixing and don't align properly, not enough to make phase cancellation work.

 

nowadays I get best results from AI,

https://www.lalal.ai/

 

Thanks but it's not what I'm looking for. I've used lalal.ai, spleeter, and although their results are remarkable, they can't achieve the level of accuracy you would achieve by having the matching acapella in your hands. What I'm looking is a program that I can feed that information to, and then remove the vocals from the full track. Ideally it would be an AI that's able to compare both audio files and find for similarities in frequency, time and dynamics, and . Any remaining artifacts could be taken care of in later processes, I suppose.

 

I know it's possible to do this without the help of a program/AI, but it is incredibly tedious work. Plus I'm not an audio engineer... But I'm more than willing to learn though!

 

You could try Focusrite Reveal

 

I will look into that one! I'll report back with the results.

 

rx-9?

 

RipX and SpectraLayers 9

 

https://mvsep.com/en/ this site has 17 methods to extract elements

 

RX9 Music Rebalance is really quite good, and maybe a little far from "perfect"; but I never strip a vocal/sample from a track and then let it play solo/unmixed. Any artifacts or results that are less than what I would call a clean pella, aren't such a big deal that way.

I'd like to see an RX9 and RipX comparison if anyone does have both.

 

In this case I want to remove the vocals as much as possible. It's worth noting that the track in question is of hip-hop genre, so transients must be preserved, and spleeter/RX9 mess up transients in sacrifice for a more successful vocal removal. However, like I said, I'm looking for a tool that will allow me to feed reference audio to it. Think of it as a sort of spectral sidechaining attenuator. Or, imagine you have a track where the first 5 seconds it's the isolated individual drum sounds, that sound exactly the same throughout the track, now say I wanna remove the drums by using the first section as reference to cancel all the other parts out. Manually this would be very tedious, so a program that I could just say "take this kick sound, then find it everytime it repeats itself in time and cancel them out". This is kinda the concept im looking after. Mix algorithm's own intellgence and raw audio information to assist and enhance the result. Hope I've explained myself properly!

 

Last edited: Aug 27, 2022

kn0ckout VST https://old.reddit.com/r/IsolatedVocals/wiki/index/tutorial01 or utagoe app

 

Basically you want really advanced De-Mixing software, when most of the De-Mixing software we currently have available are barely capable of what they are sold for. Stripping a vocal in/out isn't even perfected, imo. If I wanted to just take a drum, like your example; I would use UVI Drum Replacer and pull the kick and replacing it with a near silence.wav or some other drum. Which doesn't work very well, if you can't isolate just the drum track properly in RX/other remixing softz in the first place.

And it's still a case by case basis of how any of it works with every song you try it with. "Unpredictable quality" is generous, unless "not what you really hoped it turned out like" is your prediction. As the tools get better, less of the work is tedious and manual. But it's just less, not 0.

 

Last edited: Aug 27, 2022

Hit 'n Mix Deep Audio is a great and impressive tool, though.

 

RX9 is pretty great. If transients are a problem why not apply a transient designer after it? E.g. DS-10 by XLN

 

For now, the only demixing software that gets the job done is Hit'n'mix RipX. It's kinda processor intensive. And takes a couple of minutes, again, depending on how powerful your computer is. But has the least artifacts. Kinda magic to be honest.
Last day I wanted to know what kinda reverb was on Ariana's vocals for the song no tears left to cry. Was kinda shocked how clean it extracted every element. I know we get Acapellas to download but one can't be sure what kinda alterations have been done to that.

 

Okay, so I've been testing with it, and I've compared results with other de-mixing software/programs/AI-models, and RipX is... something else. I mean, it's incredible. It preserves the transients, it doesn't have any weird spectral artifacts, at least not ones to be noticeable. And you can even edit the stems and more, sort of like melodyne. However, the only thing I'm missing here is that you cannot "fix" whenever the program misinterprets an instrument/sound and places it in a stem that doesn't belong. There should be an option to move a selected region of sound to X stem/colour. Maybe there is such option, I haven't found it yet.

 

Once I had to remove screams and growl vocals from a deathmetal-track. It was very difficult. But RipX did the job by far with the best result!

 

Did you see my answer above? https://audiosex.pro/threads/lookin...act-vocals-in-this-fashion.66268/#post-660310

 

Oh, my bad, I must've overlooked that one. I will try your suggested plugins, I'll report back with my thoughts.