Thanks for taking the time to create that video tutorial - very helpful! It has given me a clear example and the steps how to separate 2 voices.
There are backing vocal AI models out there you can use but not in SpectraLayers. Just this week a new BVE model was released which is still beta, so being trained on x-minus web service at the moment but Iâve already used it for a project as itâs easily the best out there currently. Thereâs a little guide here too (not for BVE) if youâre an x-minus / UVR user⌠Some methods will need premium membership to use on x0minus or UVR5 though.
How to extract backing vocals with the X-Minus Premium
Obviously thereâs all the karaoke models on UVR or MVSEP too that retain backing vocals in the instrumental. Maybe an ensemble of:
- VR Arc: 6HP-Karaoke-UVR
- MDX-Net: UVR-MDX-NET Karaoke 2
- Demucs: v4 | htdemucs_ft
would be a good start? All depends on the song, but get the best vocal separation first with a main vocal model or SpectraLayers demucs and then finish getting the backing out in UVR or SpectraLayers
1 Like
First of all, great video and very nicely done. Great sound on your voice too., very clear narrative.
In your video, your process was similar to mine and in fact right down to the multiple OCD key presses. Haha. Totally with you there. 
However, I still maintain that what I said originally, is almost impossible.
In your example:
- There were only 2 vocal parts
- Both of the vocals are very clearly singing different parts, there are no BV blocks or crossover frequencies. (i.e., certain vocal parts singing in unison).
- Most importantly - The 2nd vocal was clearly panned to one side which makes it much easier to define.
And actually, in my original post, I did say that if the BV parts were panned, it would be easier to deal with.
My point was:
- 3 vocals
- All panned central
And even this was not including backing vocals which could also be central; and not necessarily panned at all.
Now when I am saying âalmostâ impossible, I have to reiterate again. That does not mean âimpossibleâ.
Perhaps this was my mistake by using a phrase that was not clear. let me phrase that another way⌠âHugely time consuming and very, very advanced for a normal user.â (I would have thought that my meaning was obvious through tbh).
And as we see from your video, even an advanced user took a long time to get a reasonable result from a very short passage that was not overly complex.
My objection was simply that you jumped in with ânot trueâ.
It would have been much simpler for you to say something like:
Yes, I agree (depending on the material) this could be a very challenging task for a new user. And may not be possible at all if the vocals are not possible to define. However, it may be possible depending on the source material.
Yes, we needed more info but by definition this means that the OPâs wishes could have been much more complex than just 2 very clearly defined vocals.
Once again, great video and I am sure the OP very much appreciates it. And so did I. I hope you now understand my point too.
And finally, what if the track had been something like Bohemian Rhapsody? Would you not agree that (although as a musician with a very good ear, I can hear all the parts, even the multiple layered unison parts), unmixing that would be almost impossible?
Haka, no need to answer that one. But thanks again for taking the time to show your process.
My very best, Phil.
Yes Phil, you said it right away in the first comment about the 3 vocal parts being all in the middle, that makes sense to me too. Maybe Steinberg is already working on a new algorithm that could automate this process, I hope. Anyway, thank you Phil. Thank you all for the engagement. âUnmixingâ made a very useful video, at least for us average Spectralayers users.
âSam_Hockingâ suggested an alternative, nice and thanks.
Greetings
1 Like
Honestly, thereâs a lot of words in here for a vocal that clearly has a defined harmony/backing so pretty easy to separate with the current models available.
I just ran this through BVE on x-minus. 2 mins and done, use the rest of the time you saved to tweak as you want to spectrally.
Video here of result playback:
BVE Harmony Split - YouTube
Stems here (if you have high rez .wav, that will separate better than the very lossy mp3 shared above)
https://drive.google.com/drive/folders/1kMzIOXi6H1MvPIMFPahpeAxyulXItaTN?usp=sharing
Respect the effort limiting yourself to doing it in SpectraLayers, but with so many harmony and backing vocal AI models out there to use I donât really see what purpose it serves? Like with normal MSS, you canât beat a suitable model that has been trained to pull harmony and backing vocals out imo.
It appears youâre presenting a question of how to extract three guys singing UNISON on a mono lead vocal. Or one guy singing in unison three times.
Or a case of how to separate McCartneyâs double-tracked, mono lead vocal on Beatles âWe Can Work It Outâ into two distinct individual tracks.
Or Three Dog Night singing in unison the melody verse lines of âLiarââŚsplitting the mono to three individual tracksâŚ
Cory, Chuck, Danny.
That type of extraction capability isnât as important to me personally at the moment as getting those two darned Strats separated with one button from âDay Tripperâ 
Yeah! About the Beatles:
I personally have observed/noticed that most of their mixes has a lot of chorus-y flanger effects which gives off the appearance/illusion that more voices/instruments are stacked on top of each other. I dont believe that is 2 separate strat guitars in âday tripperâ, I believe there is an acoustic guitar and a strat guitar with heavy flanger or chorus-y effects (especially on the high end). Technically it is possibly to unmix the effects (of reverb, delay, choruses, flangers) but as far unmixing a source/sound that isnt there isnt practical.
The solo consists of two strats, both overdubbed at the same timeto a single mono track of the four track. These would be knocked down a generation (along with drums, acoustic guitar) during the later submix (reduction mix) over to the second machine (new reel of tape) to open up tracks for the two passes of background harmonies and double tracked lead vocals.
Solo begins on BâŚ
John continually plays the main riff on his strat.
GeorgeâŚat beat 1âŚon his StratâŚhits lower strings of B chordâŚlets it ring until adding an ascending single note (c# is the first of the upward series), anticipated just before beat 6.
George continues with the B chord hit+ ascending singular notes against Johnâs continuing riff.
At bar 7, John transfers to simply hitting staccato 8th note lower-strings B chord while George transfers to playing the bendy note lead solo (where the ahhs will eventually be overdubbed after the reduction mix).
At just before the bendy section starts, you can hear/detect Norman raise the fader on George.
You can also hear the effect of John & George playing the main riff on each strat all the way at the beginning of the track. Which they remain doing until the solo section where each guy branches into the two approaches above.
Video seems to be disappeared, can you post it again? I´m very interested in different methods
Maybe I am missing the point here, but wouldnât this be better done with something like Melodyne.
Use SL to separate all the vocals from the instruments, then use a DNA polyphonic version of Melodyne (Melodyne Studio) to separate out the individual vocal parts. Iâm assuming here that each vocal you want to separate is on a different pitch simultaneously⌠which might be doable. Separating different voices singing the same pitch would be black magic and almost certainly beyond possibility.
Nope! Melodyne operates on the same principles of harmonics and overtones. The only difference between something like Melodyne is that it seems like they built their own custom library to detect sources and they define those sources as the fundamental + the overtones series that follows. Whereas something like Spectralayers is a map of all the audio details and you can use tools (such as the harmonics tool) to select frequencies and tones. I use to own Melodyne years ago and I havenât used it in years so I am not sure how advanced they have improved from when I last used it.
Which is why I am standing by my original statement when I say that The Beatles record âwe can work it outâ is either one vocal that is duplicated to sound like a overdub or a chorus/flanger effect. Even on most overdub recorded vocals, the overtones/harmonics profile (9 times out of 10) is distinguishable in frequency and time.
I unmixed the project file here and would like all the experts here to download the project file and tell me if what I am saying is true. I challenged @DosWasBest and even invited @DosWasBest to download the unmixed project file and so far didnât here anything back from that user
Interesting. Your spectral editing knowledge is beyond mine. I do use SL and find it very useful for fixing problems and remixing, butI hope to learn much more about it as time goes by.
Iâve downloaded your project and will be taking a look soon. Thanks for taking the time.
1 Like
Just having a first listen to your unmix project. The track labelled Strat doesnât sound like a guitar - more like a keyboard or melotron.
1 Like
@SIedDriver Thank you! which is what I first initially thought as well. The only reason why I named it âstratâ was because of the fact that @DosWasBest said it was a âstratâ and then I assumed there was some type of article(or publication) out there of âThe Beatlesâ recording that record where it details that there was indeed a âstratâ guitar involved in the recording process.
The âStratâ track is primarily harmonium. Itâs well under in the verses and more prominent in the bridges, but is there throughout. It was played by Mr. Lennon.
McCartneyâs lead vocals are double-tracked.
Chewy
Source? Any links(to any publications of the recording process) I can view to verify this?
Iâd refer you to Mark Lewisohnâs âThe Complete Beatles Recording Sessions.â Iâm on the road now, so canât, for whatever itâs worth, give you page references. But if you want an easy confirmation of the harmonium thing, outside of "common knowledge for Beatles freaks 
", Iâd say just do a quick google search for âwho played harmonium on âWe Can Work it Outââ, or search for the song wiki itself. Thereâs lots of detailed info on the tune on Wikipedia, much of it clearly sourced from the definitive Lewisohn documentation. In regards to the double tracked Vox-- I couldnât tell you if it was ADT or actual double-tracking, but Paulâs ability to duplicate his own phrasing shouldnât be underestimated.
Chewy
If there is indeed a double vocal overdub track(which I highly doubt), then I already unmixed it. In my unmixed project file I ended up with 3 main vocal sources for the lead vocal. There is dry vocal and then another dry vocal and then a vocal source that appears as the wet reverb or chorus/flanger effect. So there are 3 sources for the main vocal in my project file.
About the harmonium: To me it sounds like a Accordian (but thatâs just my opinion).
However I unmixed the Acoustic guitar from the harmonium.