I see where SpectraLayers Pro 11 is supposed to be able to unmix voices and it is shown in videos. I have a song track with a duet of 2 lead vocals and a chorus section of 4 vocalists or vocals. I can unmix the vocals once. But I want to do something similar to “Unmixing Multiple Voices”., but for vocals. First I would like to separate the duet lead vox parts. The 2 lead singers sing individually and together in spots. - Is this possible and how? I have tried many of the unmix modes and most end up giving me “Can’t unmix a group.” Do you know the reason for this? If what I talk about above is all that can be done. I am ok with that. The icing on the cake would be if it is possible to unmix the 4 vocalist chorus sections. - Any ideas on that? - Thanks!
You can’t select a group and try to unmix.
If you already have unmixed the vocal layer and wish to unmix further you must first select one of the new layers created within that Vocal group.
It’s a hierarchy where one group can contain layers and/or groups (that can contain layers and/or groups, and so on).
Thanks Marshall,
That adds clarity and makes sense!
On another note I think I am onto how to split out vocals the way I outlined in the OP. If I come up with something solid I will post the method.
LB
I made some progress by watching a video on vox unmixing that can be found here:
The video should take your right to the section on “Unmix Chorus”.
A top level view of the steps I took were as follows:
- Unmix Song module to produce the full vocals track.
- Unmix Chorus was ran next and did fair job of isolating the 2 lead vocals in the duet from the backing chorus. But I deem the results unusable in quality.
- I tried Unmix Chorus and Unmix Multiple Voices to separate the 2 lead male/female duet parts. This came out very poor. There were dropouts, artifacts and more in the results.
I did not come close to the nice results that were in the video. But he was also not trying to split out duet parts.
My thoughts on all this are that maybe my methods are wrong? If that is not the case I am left with SL11 not being ready for prime time on this type of operations. I would like to see a new extraction preset module for vocals that works similar to Unmix Multiple Voices. Whereby you can tag a vocal segment and the app uses that to separate out the individual voice tracks, like one would see in a duet.
If anyone else has and ideas on this, please chime in!
LB
Unmix Chorus is really for Lead vs Backing vocals.
Unmix Multiple Voices is designed for speech separation at this point, not for singing voice separation. That being said, it might technically be possible to upgrade it to singing voice separation, but that requires more R&D. As it’s, it’s rather for separating 2 voices in a podcast or interview for instance.
Thanks Robin!
Yes, pretty much what you said is what I figured. I hope in the future, more consideration is given to vocals and separating them out to stems. It would be truly an incredible tool for my work if SL could do this.
LB
I’ve never used spectralayers before. I bough 11 pro. I tried the above to separate a male and a female vocal. Tried chorus and voice unmix. Neither worked well. I wondered about other options. Can you separate using vocal pitch ? Can you sensibly go through a track pick vocals out. Remember I’ve never used this before so I know virtually nothing !! M
Hi @Yeastieboy, the Unmix Chorus module is meant for unmixing lead vocals from backing vocals, and Unmix Multiple Voices is meant to unmix speech, not singing.
To separate 2 singers male/female you can try the following :
- the Voice DeNoise module has a “Other Voice” background mode, meant to remove one of the two voices in case a voice is heard over another one. See if that helps removing one of the 2 voices. If so, then you can first duplicate your layer, then apply the module on the first, and apply the module on the second with the Reduce Signal mode this time (reverse effect).
- or you can do manually with the harmonics selection tool, which is a tool you hover a fundamental frequency or one of its harmonics, and then click and drag and it’ll automatically follow how the pitch change to keep selecting the same content through time. I’ve recorded a brief video about it today in this other topic : Separation of lead guitar track? - #8 by Robin_Lobel
Because @Robin_Lobel is suggesting that it can be done using the tools provided by Spectralayers here, I want to also point out that I was able to do it HERE on a very compressed video with all 5 lead singers being center in the front using Spectralayers pro 9. Knowing what I know now, if I had to go back in time and redo this unmixing I would rebalance the NOISE levels better between all the vocals a little more (because if you notice the first 2 vocals is deprived of acoustics and room NOISE whereas the 3rd vocal has most of the acoustics/room NOISE).
Thanks very much much. I’ll try these options and get back to you . Cheers Mark
As an aside, once isolated mono stacked harmonies are away from main instrument tracks (which SL does very very well for most source material I work with)…I manually separate harmonies from each other with various other programs…placing each voice then on a separate audio track.
While a bit more time-consuming for a 4-minute song, it is a very efficient approach to getting clean separations…plus…no chance of additional artifacts being introduced.
For anyone proficient at microscopically hearing each harmony in a stack (particularly for parts that weave amongst each other), manual separation of the vocals is soooo nice now that the historical roadblock of first getting vocals away from instruments has been solved over the past 4 1/2 years.
Speaking of stacked vocals and demixing…we’ll all see how Giles/Peter Jackson are polishing their demix skills on the next Beatles 64 doc, releasing on Disney in a couple of weeks…Lots of stacked mono voices on that source material…we’ll see if that crew does any splitting of the three vocalists parts themselves.
I am very interested in this, and have practiced in SLP for hours…
Can you please share which ‘various other programs’?
He probably means UVR and alternative models
Dunno what uvr is.
After the initial steps to do a general separation of a track full of instruments/vocals (particularly mono and usually with Sl if that algo seems to be working best on a track), Melodyne studio (in conjunction then with followup chains is one astronomically fast manual way (for me) to pull stacked vocals apart, export each and then continue on with three other processes I use before re-importing the individual mono vocals into the overall Cubendo projects. I don’t always use Melodyne, but I often first load the stack vocals in to visually see what parts are crossing each other when/where. Then decide a course.
Harmonies in straight thirds are easy. But when you get into harmonies where, say, three singers are RADICALLY moving around in a weave fashion, crisscrossing each others’ ranges all-the-time for four minutes…well then it’s much more effective for me to visually spot the movement WHILE I’m listening. This lets me also consider if tonality is similar/dissimilar between vocalists…and best way to pull apart…all part of the manual process for me.
I tend to begin with SL as an initial demix step. Much later in processes on split-out components, I may do various amounts of sl nr on various clips, but only very cautiously as nr is actually kind of evil for restoration projects when used heavy-handedly 
My tendency with Cedar systems in the 90s was “hey, nr is great and a revolutionary tool”. Over time and zillions of projects, I understood that …no…nr can completely ruin things…but I digress…
At any one time, I’m working on over 50 client projects. They’re each radically different from each other in sonics and each require varying approaches.
I do often also send snippets of things to various online outfits (once I determine they’re not spleeter based) just to gauge developments various coders may/may not be making in…say…pulling a kazoo solo out of a mono B3/lp Marshall stack mix
Also…my personal mindset mantra for excellence with project results is …never ever ever ever use ara “anything”. This type of work takes focus, compartmentalization of workflow, and ability to seamlessly switch tools for each step. My opinion after working with thousands of multitracks over the decades.
One thing at a time…no chained-together processes…and absolutely no ara! I sound like a broken record…but I can fix that
Ultimate Vocal Remover, theres discord channel called Audio Separation, where you can find additional info and models, even datasets and info for training models by yourself, really helpful folks there
Oh, yeah, I downloaded/tried that last year. Worthless for what I do…but like I say, I try stuff that’s out there just in case.
I’ve tried Melodyne Studio, to no avail.
Have tried both; -to adjust the volume down, -mute and -delete harmonic blobs. I didn’t get any of this to work satisfactorily…
Could you please tell me how you got on with Melodyne; step by step?
It would really be a great help.
Thanks
Hi, if you could spare some time and give us an outline of your Melodyne procedure, that would be highly appreciated, thanks. You certainly have an impressive amount of experience.
Hi . I have tried some options and it hasn’t really solved the issue but now I have another!! I managed to unfix guitar and drums well so am trying to record a new (hopefully better) guitar track. The problem is that the new track is never in synch with the unmixed tracks. Sounds fine as I’m recording but on playback they are different. Can you tell me why this might be? Thanks again Mark
I suppose the new guitar sounds a bit later than the other layers because of the system’s latency.
You can somewhat fix this by selecting all of that new guitar and move it slightly to the left with the Transform tool.
Not an elegant solution but it’s doable.