Bass Isolation is Still Terrible in SL-11

SpectraLayers
,
1

I had high hopes that SpectraLayers 11 Pro would have improvements over SL-10 for bass extraction and isolation. These hopes were unfortunately dashed right out of the box.

There is virtually no difference between bass extraction results when using SL-11 Pro versus SL-10 Pro. Worse still, SL-8 Pro and SL-9 Pro both exhibit far better results when extracting and isolating bass.

Much of the bass layer for many songs is lost and remains mixed with the guitar layer. Even songs with very prominent bass have large sections where the bass suddenly and inexplicably disappears when isolated to a layer. Activating the guitar layer along with the bass layer magically fills in the missing bass sections. The only recourse is to manually extract residual bass sections into new layers, and then merge them into the original lack-luster bass layer. Even when set to “EXTREME” Unmixing, bass extraction is horrible. What a huge disappointment!

I am forced to use SL-8 Pro to first unmix, then SL-9 Pro to further unmix and isolate bass notes. I then have to import-as-layers those respective bass layers into SL-11 Pro, and merge them into the failed bass layer within SL-11 Pro.

Here are excerpts from the “live Event” on YouTube:
11:18 AM: Dom Sigalas - Everything is improved :slightly_smiling_face:
11:19 AM: Precision Productions - ​​How about the bass isolation, Dom? Is it improved over v10??
11:20 AM: Dom Sigalas - @Precision Productions try 11 :wink:
11:20 AM: Precision Productions - ​​I can’t wait!
11:21 AM: Precision Productions - ​​This version looks awesome!
11:22 AM: Precision Productions - ​​I got a full-on SpectraChubby!

Well, that certainly went flaccid fast when I found no change or improvement in overall bass isolation and extraction when compared to results in SL-10 Pro for the very same song unmixes.

Much of SL-11 Pro is outstanding, and a marvel of modern software, but I think Steinberg should realize that “Vocals” do not a song make. Sure, it’s great to have awesome vocal extraction, isolation and separation tools, but how about a little love for bass?

Guitar too, needs improvement, as oftentimes hints, or even large portions of keyboard sections remain peppered throughout guitar layers. Again, it’s not all about “Noisy Speech”, Steinberg.

It would be truly awesome if we could have an option to “Unmix Guitars” in the very same fashion as “Unmix Drums” functions. Much of the content I unmix contains multiple guitars. Dare I dream?

This is incredible software, and I am a very firm fan of its performance, as well as an extremely loyal customer, not only regarding SpectraLayers 11 Pro, but also Cubase 13 Pro, Nuendo 13, The Absolute Collection and the UR-C series of Audio Interfaces, all of which I have purchased and own. I really admire the way Steinberg does business. I admire their customer support. I really like the way they go out of their way to provide useful and “cool” new things for free to their customers who have invested large amounts of money into their company and its excellent software/hardware. I also very much appreciate this forum where I can vent my frustrations, though they are very few. However, here comes the “but”:

BUT, I really think that SpectraLayers could stand significant improvement in the isolation of all manner of instrumentation. Drum isolation, drum unmixing and overall drum layer performance is awesome. Why can’t we have all that for guitar? For bass? For Keys? I might add, that in most instances there is not just “Piano” there should be a “Keys” category at the very least. I like the new Brass categories too, but they seem to just end up with keys and organ residuals, especially when no brass is present in a song.

I noticed that "Non-Unmixed has disappeared with little to no fanfare. I think that’s a good thing, because it was a bad idea right out of the gate. “Other” was plenty. I would often end up unmixing “Non-Unmixed” a second time, then doing the same to the “Other” category. I would then merge those secondary layers back into the original layers where they belong (ie. drums to drums, bass to bass, etc.)

Please, Steinberg, FIX bass isolation! Test it yourselves against SL-8 Pro and SL-9 Pro; you’ll see precisely what I mean.

In addition, please IMPROVE isolation and separation of guitars, keys, and bass. Often, upper frequencies of bass remain in guitar layers. I realize these instruments are very close in frequencies, and that what I am asking may be a tall order in some cases, but if the human ear can differentiate these various sections of a song, so should Steinberg “AI”. After all, have we not been warned that “AI” is going to swallow up the entire world? (lol).

This went a little long, and might possibly come across as a bit of a rant, but I feel very passionate about requesting improvements in the areas I have outlined here. Thanks again, Steinberg, for providing this forum where I can be heard (read), and have my say.

If anyone else in the community agrees with me, please make yourselves known in this thread. We need as much “squeaky wheel” as we can get here, so we can get “the grease” in the next release! Dare I hope that such improvements could actually be released in a patch/revision?? That, would be amazing!

Thank you for reading!

5 Likes
2

I am as big a fan as PrecisionProductions and as loyal too, going back to 1990-91 and Cubeat. I have also been active in the team that developed the Virtual Guitarist and Groove Agent plugins some 23 years ago FWIW. Have all the big Steinberg apps except for Dorico (I don’t read music). I think we are many very loyal fans here.

I agree on most points!

I really love SpectraLayers too, it is a brilliant piece of software.
The new SL11 is marvelous but I’m a afraid it feels like it’s not quite ready for public exposure yet; there are some serious quirks. In previous versions such problems have been ironed out with the first one or two patches that followed and I’m confident this will happen with SL11 as well.

When it comes to the Unmix and separating features all in all they are way better than in SL10; the resulting stems sound clearer and more powerful. That goes not only for Vocal but also for the other stems.
Even the bass separation results are better in SL11 in my opinion. It’s quite noticeable when working with well mixed/mastered material in 24/96.
But it could of course be better, it’s not perfect, like Drums and Vocal almost are now.
The ”Extreme” setting in ”Unmix song” makes killer results but takes very long time and unfortunately there’s not much of a GPU exploiting on a silicon Mac.
But hey, there’s the new batch processing feature (and modules chain) which makes the ”Extreme” setting usable anyway. Hooray for that!

Hoping the first patch won’t be to long. :pray:t2:

3

Bass can be really hard, just try to separate bass from Pan Ron - Hope to meet you

4

Wait a sec op…what is this “the bass” you speak of. Is there only one bass guitar on the planet? And it’s on every recorded track known to mankind and “it” can not be successfully extracted?

You hint that ALL bass guitars suck when extracted with SL11.

Just for context exampling…run the stereo version of Sugarloaf’s “Green Eyed Lady” through sl11 and unmix it…is that picked, direct Fender bass sucky?

I’m not saying it is or isn’t…just wanna know your definition of suck…as you didn’t post anything of yours that you’ve extracted.

I get wonderfully extracted bass guitars from wildly-differently-engineered stuff (my own) that was recorded over the decades in different ways in different studios…often using multiple approaches/programs to get the clean bass gtr extractions…but it can take work for me…

Sounds like you just wanna push a single button on your refrigerator and always have clean, refreshing bass gtr flow into your cup without thinking too much about it :slight_smile:

2 Likes
5

Yes I’m a bit curious about this specific bass sound @PrecisionProductions is referring to. Because on the wide variety of songs featured here, bass are properly extracted - and with better quality than SL10 :
https://divideconcept.github.io/Unmix-Comparison/

Also keep in mind that AI is just an accelerator - SL11 provides several tools to let you do manual bass extraction yourself if AI fails, or refine results from the AI.

1 Like
6

Hey,
When I try SL to separate bass stuff, it really never works, because I want to separate plug synth bass stuff, which is hard, since you have the “normal” bas stuff, but also a lot of mid stuff (attack) for dance / house music, that makes the bass sound complete. I don’t think that SL knows how to make the whole complete bass sound , as SL would need to know about sub / bass and attach for many synth basses.

But the new transfer to layer… feature, could really help on these sounds, as the click, plug of the synthbass in places in other layers, that might not be too crowded, and therefor easy to transfer to attach of synthbasses to the bass layer.

For real bass guitar sounds, it might / should be a lot easier.

SLL

7

I agree in some circumstances the bass extraction is not so good - but it is a lot better in SL 11. This is not surprising to me as I am processing a lot of live tracks recorded with a pair of stereo mics. If you consider that a bass guitar is not really very different from an electric guitar - especially where the lower notes overlap in pitch - this is not at all surprising that any AI stem extraction algorithm struggles…
You can see this clearly when you unmix the whole song - there will often be large sections in the Bass layer where all the bass is missing and has gone into the Guitar layer.

I think the way SL works is to sequentially extract each layer - first the Vocals then from what is left extract the Drums, then from what is left - the guitar etc. So knowing this you can sometimes get better layers by doing one at a time. Yes it takes longer, and yes the sum of all the layers may be greater than the original track (nothing is lost but additional information might be in each layer). I think the new batch processing with an appropriate module stack will do this easily and when the GPU capability is fixed then this should be fairly painless…

So what I do to get a better bass layer, is to first just extract the bass layer ONLY using the Unmix Song module (i.e. with just the Bass checked). You can then run the Unmix Song module again on just the Other layer and extract the rest - guitar vocal drums piano etc. You can then balance the Bass layer to taste with the rest of the layers (perhaps also cutting many of the low frequencies in the Guitar layer).

This works well for me .

2 Likes
8

Actually! The new transfer tool seems kind of gimmicky to me and it seems like it was added subtlety as a “new feature” in order to lure customers into upgrading to Spectralayers. The new transfer tool could’ve easily been implemented within the subset toolset options (with the ability to choose which layer to transfer to)

20240624_133255
20240624_1332551920Ă—1080 128 KB

20240624_133530
20240624_1335301920Ă—394 80.7 KB

But I guess I kind of get why it was added as some sort of “new feature”, I guess when you’re trying to hit a certain threshold of features to add (maybe there was pressure from the Steinberg to add unnecessary features) then it might cause you to try to add things that are not necessarily needed in order to woo people into believing this is something totally brand new that is useful and needed. It kind of reminds me of the presentation apple and AMD and Microsoft recently gave where they’re trying to convince people of how impressive A.I. is by convincing the public that “A.I. is so futuristic, don’t you see how cool this is”. It’s like they’re trying to woo people into believing A.I. assistance talking to each other is so impressive and actually useful in everyday life. If you pay attention, you notice that most of these new A.I. tools (from Apple) don’t really solve real-world problems (like taking a bone/fragment from a dinosaur or any corpse and accurately reconstructing what it would look like in real life).

The magic wand tool (along with shift + X) is the best solution for this because most basses have an attack (Amplitude/Volume along the vertical axis) and the magic wand tool is intelligent enough to detect what is on the vertical axis along with what is on the horizontal axis.

9

The bass isolation will always be appalling.

10

I too have struggled to get as good results on bass with SL11 as I did using SL9, with the bass ending up conflated with the guitars layer. Piano is sometimes the other problem child. Undoubtedly, the track I am working with now (Derek & the Dominoes “Key to the Highway”) presents challenges, as the piano, bass, & guitar layers all overlap significantly below 400 Hz, and of course, the recording itself was done many years ago & subsequently remastered digitally.

11

EDIT: Apologies: I meant to post this in a different thread, but I’ll leave it here also, because it’s somewhat relevant to this thread as well.

BIG “SpectraLayers 12” wish list for me would be an elevation of instrument isolation to the level of vocal isolation/separation and drum isolation/separation. Namely, separating multiple guitar tracks; better bass isolation and separation of bass upper frequencies from guitar sections; less keyboard-bleeding into guitar sections and vice versa, and a LOT less leftovers in the “other” category that have to be manually separated and then redistributed among various other stems.

In many cases, I have found that versions 8 and 9 are far better at isolating lower frequency bass-guitar sections than either version 10 or version 11, which sometimes leave all bass-guitar sections completely un-isolated, or more often leave huge chunks of bass-guitar “missing” for no apparent reason.

Every new release of SpectraLayers is a leap forward in technology and usefulness, and I always look forward to SL releases with much more excitement than any other Steinberg product that I own, such as Nuendo, Cubase Pro, The Absolute Collection, etc.

I can’t wait to see what amazing developments are to come for SpectraLayers.

1 Like