Spectral Layers 11 Doesn't Completely Remove Drum Artificats

SpectraLayers
1

I had spectralayers 10 and was more or less happy with the way it unmixed a song into stems. I use these stems to assist me with transcription, sometimes with automated AI tools if I can get it clean enough.

The first thing I noticed with Spectralayers 11 is that it did, as usual, a fantastic job of removing drums, bass, and vocals. Yes,it did catch the odd sax or rhodes piece as a part of the vocals, but that was minor.

No drum artificats in the vocals at all usually. But everywhere else? It’s all pervasive. And it is a pain in the ass to remove those artifacts which, if memory serves, I didn’t see so much of in V10.

What can I do to minimize this so I’m not wasting HOURS trying to remove them.

Also, while I appreciate the Sax & horns addition, it’s not really ready for primetime and seems to grab all kinds of things to go there.

What is Absolutely needed is STRINGS remove and some subtle tool that allows you to differentiate between an ORGAN, FENDER RHODES or WURLI, or PIANO. There’s a tendency to take the obvious strong piano parts and put them there. But everything else goes in OTHER.

Don’t take this as bitching at you. The work is stellar.

But what are some things I can do to automate fixing these little pains in the ass. Can I make automated module presets that will do most of the dirty work?

Honestly, though, those percussion artifacts are driving me crazy and killing the AI tools that are doing transcription for me.

Thanks in advance. Please don’t flame me.

1 Like
2

Oh, quick point. I’m aware of the extreme difficulty in differentiating between those keyboard instruments I mentioned given to the frequency range they all cover. I know you’re not magicians.

1 Like
3

Hard work and dedication. Keep working and put in the long hours!

See! I knew this was going to happen. This is going to turn into a never-ending request of instruments to be added to these algorithms. Just the other day someone asked for different car models (like Ford and Toyota) to be added as algorithms :smiley:.

Had this been done right wat the first time, this could’ve saved the developer tremendous amount of time and energy.

2 Likes
4

Yo, I was totally joking when I wrote that

5

Certainly SL 11 is a progress compared SL10 but the un-mix doesn’t reach the objective.
Even drums and bass almost un-mix on each channel contain frequencies of other instruments and noises.
Good luck to clean them (hours (days) working without good result)
I would appreciate if SL 11 un-mix really all instruments and in the OTHER will contain only the noises which can be clean just deleting .

Instead to use your file, I would appreciate also if you can un-mix this song which has been a nightmare for me
P.S. This song is mine and I can remove the copyright

1 Like
6

@Gennaro

I would love to help but…

I listened to your file and that sounds super uneasy to unmix however I know that if I ask you to send me the high quality version of this and I unmix it then I’ll have trouble sending this because If I unmix this and send you the raw uncompressed file then I can guarantee that the file size will exceed 10GBs.

1 Like
7

we use wetransfer for large files…full res video can be pretty weighty

8

@Gennaro
This is a really beautiful tune…

9

Thank you so much Joey
I was just looking for to use this song showing the ability on de-mixing a classical music of SL11.
However, being my composition I have all files in *.mid and *.wav already.
Actually I did also style and multi pad on my Genos to play it live
Maybe I am asking to much from SL11
What is clear that some frequencies are omitted on de-mix process.
Now I don’t know if something can be done to improve this side of SL

1 Like
10

Unmixing is essentially far more difficult than mixing.
Just think of mixing colors together. It’s fairly easy to do. But unmixing those colors…can become a very challenging task. It’s like a chemical decomposition of something unknown.
What AI does today with unmixing seems like a miracle to me, but it’s still in its infants, imho. I wonder, where we will be in 5, 10 or 20 years from now.

1 Like
11

@ Marc_von_Bredow

Yes - the process is still only in its infancy…
It’s a shame to say it, but I think in five years you will find un-mixing apps - with better algorithms, training grounds and more instruments (almost) for free on all mobile phones…

1 Like
12

There’s always someone out there willing to demonstrate how incredibly fun it is to unmix and the best thing is these instructors/teachers don’t even use none of the algorithms to unmix. If only someone(Steinberg) could sponsor a masterclass for Spectralayers.

2 Likes
13

It´s also how you synthetize the sound back, not only models… it´s all digital anyway

14

I’d be willing to give you an assist as well. As the prior poster mentioned, lovely tune. Irish? Celtic?

If you can set up somplace I can download a high version of it as well. . . Glad to try.

Which stem in particular are you trying to isolate?

15

Oh yeah, no doubt. Vocals, Bass, and Drums are generally easy to unmix because they lie in narrow frequency ranges. All the stuff I was bitching about above? It’s not really so much of bitch as I am absolutely aware of how much dynamic range any keyboard has, as do strings (all those harmonics).

There are so many new features, I haven’t had a chance to really try them yet. For instance, mid vs side, vs channels in combination with unmixing. It almost takes more time to unmix with a semi-clean mix than it does to produce a cut.

BTW, has anyone had any luck in using EQ and compression before an unmix to tease out a particular track? For instance, let’s say I have an organ and piano on the same song. They both occupy more or less a really broad range. I’ve tried to use EQ and shape that curve to bring out more of the instrument I want, but really am not good at it.

I’ve also found that highlighting a section that repeats from a particular interest and then “select next” or “select all” really isn’t working for me the way that it does in the video tutorials I’ve watched. I keep wondering what I’m doing wrong.

Ideas? Beyond glib answers? LOL

16

In my experience unmixing a compressor treated file - that is with squashed dynamics - works worse than unmixing a file with good dynamics.
Possibly it will work better if you can apply some ’dynamics bettermaker’ before unmixing a song. Like Zynaptiq’s Unchirp, that can enhance transients.

As for using eq pre unmixing to facilitate the process; yes, as long as the eq:ing makes dynamics better it will facilitate unmixing.
And applying more bass before performing Unmix song seems to make it easier for SL to identify and separate the bass instrument from the others in my experience.

17

Actually, that’s cool!

I had another thought. SL does it’s own compressing as it splits the stems. In fact, like all tools of the sort, each stem is really throwing away a part of the frequency.

Would this also be useful for running over the exported stem?

18

this is news to me…where can I read up on this and what kind of “compression” does SL add? I mean, I don’t hear any compression, per se

19

No, there’s no compression going on when SL unmixes into stems.
The unmixing processes are nondestructive, meaning that the stems played together will null with the original un-unmixed file.

As for using a transient enhancing plugin like Zynaptics Unchirp on the resulting stems, one is always free to try. I don’t really see any specific meaning generally though.
But I often use their Intensity plugin, to beef up drum stems and make them sound clearer.

2 Likes
20

Here is an example of my own custom algorithm I’ve been working on for years, it’s not based any of those other algorithms that are based on the average mean sum of stems(Bass/Vocals/Misc/Drums). Non of the other algorithms could unmix this because the timbre(Pulse Wave, Square wave) of each sound is closely related… The audio is “Pokémon National Park Theme”.

1 Like