Just curious - unmixing (live/multitrack) orchestra performance

Hi -

SUMMMARY - are there other ‘sound separation’ tools out there, focused on (algorithms trained by) orchestral instruments/performances.? As opposed to usual, Song, Drums, Guitar, Bass, Keys, Vocals, etc…

I understand the available selections of ‘algorithms for unmixing’ can be tried on all kinds of material; there’s no restrictions. Results will vary, according to (the quality/complexity of) the source.

However, maybe I’m missing something. Does anyone know of other specific tools designed with (trained on) orchestra instruments/setups/performances, from the outset.? Classical, film score, or otherwise (Big Band/Swing, tribute concerts of famous pop/rock artists catalogue, etc…).?

Am curious because I have a bit of work coming up where I’ll be able to try things for myself. It’ll be my first time at looking into the world of orchestral audio performances - mixing, editing. And yes, I’m expecting that will be a whole different story in and of itself.! :slightly_smiling_face:

Its a community led ‘project’ orchestra and am interested, if I might need at some point to examine ‘unmixing’ for example, any brass/ww bleed into the strings mics, etc…

(EDIT - In talks with the recording engineer, he has already indicated, that at previous concerts this has been a serious concern, when he test listened back to captures of individual mic feeds, afterwards. Issues are mostly due to size of the stage and proximity of players - neither of which he has control over of course; positionally speaking, that’s down to the conductor/players preference).

My thinking right now is I shouldn’t aim to ‘fiddle too much’. But, I know I’ll be wanting to give it a go - because, I can.!

Anyway, in the meantime I’d welcome others views - specifically here, about using SpectraLayers folks.! Bigger picture concerns/topics/discourse, are for elsewhere… Many thanks.

1 Like

As I mentioned before, I was in the process of building my own application (it wasn’t based off the typical 4-5 outputs of bussed/summed outputs but based off of my own unique approach) but I bit off more than I could chew. Then on top of that, I am currently going through a lot of DRAMA in personal life, not looking for a sob story but I am dealing with multiple lawsuits and between the court cases and litigation that has taken up the majority of my time.


If you can, could you send me the orchestral performance you would like unmixed. Anything can pretty much be unmixed the only thing that might be concerning is multiple layers of instruments playing in time at the same frequencies. Although can still be unmixed, multiple layers of instruments(for example 4 violins playing the exact same thing at the same frequencies) piled on top of each other can be extremely challenging because you can hardly tell the difference between room noise(convolution room reverb) and the instrument itself.


Thanks @Unmixing - sorry to hear of your personal ‘troubles’ right now; hope all works out ok for you…

As I said, I was more curious about the ‘trained algorithms’ out there, for these sound separation tools in the market and whether any had an orchestral instruments focus, specifically. I guess the training would be a far more difficult thing, on string ‘sections’, brass ‘sections’, woodwind ‘sections’, etc…

I have had a brief look around on the web and yet to see any…

Of course, I will try the tools SL gives first… (I recently upgraded to ‘Pro’).

More broadly, I will need to be patient for the project to start, as to when I’ll be sent the actual audio multi-tracks for loading into my DAW (Cubase). I can’t properly assess anything until then, which will dictate what path I eventually take to get the job done - some stuff I may not mess with too much at all; other parts I may just want to try and see if ‘unmixing’ will bring any benefits/clarity.

I think you’re right in one sense, it will be several violins layered on top of each other, playing at the same time around all the same frequencies, etc… and more besides. From my brief conversation with the recording engineer a while back, it will be an issue of dealing with ‘bleed’ onto the strings mics (violins, viola, cellos) from most everything else.!

Anyway, no knowing until I actually hear the tracks. Thanks for now.

1 Like

Actually no! Surprisingly the only difficult thing is processing power.

Yeah sorry I still haven’t figured that one out yet. The only way I can kinda sort-of see this being possible is if you completely dry(completely DEPRIVE of all reverb and room noise) the audio signal (and mind you! This is also taking into consideration under extreme circumstances that the same instruments are spaced out across the room). However (as we all know) orchestral performances are literally based off of the concept of reverb.

If it’s a violin and another instrument like the cello then that is doable to unmix (because even though they are similar in the sound profile it’s 2 different timbres). It’s easy to unmix a trumpet from a saxophone because (although similar in sound/voice profile) it’s 2 different timbres. However if 2(or 3 or more) of the same instruments is playing at the same time in the same frequencies then I wouldn’t know how to unmix that (or if it’s even possible to unmix) (because it’s registered and viewed as one source.

Yeah! Like I mentioned, if you remember can you please send it to me (or at least a short demo snippet). I like to experiment with stuff like this.

Hi Unmixing,
You say"It’s easy to unmix a trumpet from a saxophone because (although similar in sound/voice profile) it’s 2 different timbres".
Really? Is that with SL or another piece of software? Could you possibly describe your procedure? Thanks Al.


Yes! In that context I was replying to. I have done it before.

A combination of my own algorithms(not based on the conventional stem formats that are known) along with other applications and research papers.

Letting your ears rest and coming back to a project with fresh ears. I find that when unmixing that it is incredibly easy for your ears to fall prey to oddities and artifacts especially after long periods of audio playback. The most destructive thing about your ears falling prey to oddities/artifacts is that it is so subtle that you literally dont hear it. Taking a break (maybe 2-3 days) is necessary for a nearly accurate judgement of audio assessment.

Hi Unmixing and thanks for taking the time to reply.
You advice about fresh ears is very sound, I’ll confess I’m not very good with that kind of discipline…
As far as software goes, clearly my knowledge is limited to reading manuals, I can’t make much out of research papers.
I own a copy of Rip DAW Pro and also Ultimate Voice Remover UVR5, which has an isolate horns that I find useful, but that won’t seperate Chet Baker from Gerry Mulligan :slight_smile: Thanks, Al.

1 Like

No…there are no trained algos in the github etc code groups… in the form of code presets…by anyone…for separating a mixed tuba from french horn from clarinet from on&on&on.

There are non-coder spectral-based guys who will attempt a demix of a 20-minute piece like “1812 Overture” through existing tools…for sure no doubt for hefty fees and uncertain results.

But no, Billy Nobody over in Detroit can’t buy xyz program…load in 1812 Overture, hit demix…and see a resulting 50-line list of demixed layers.

However…it’s early in the century. We have 70-some more years before the year 2100.

The preset technology will certainly get there eventually :slight_smile:

1 Like