Spectralayers 12 issues

Exact same issue here. Testing some electronic music. Enabled Sax and Brass (song has none) and enabled Unmix Sax and Brass first. In High mode, I was actually able to get some sound on the Sax and Brass track while the others were empty, so suspect it’s something to do with this new ‘feature’.

I ‘fixed’ this by setting decoding/rendering using CPU only. Now getting drums, vocals etc using the High setting.

CPU: Ryzen 9900X
GPU: RX 9070 XT
Windows 11 Pro
Tried WASAPI and ASIO - same results. Tested in Cubase 14 - same results. Latest AMD Drivers and Windows Updates.

Spectralayers 11 used around 50% GPU, where v12 uses a lot less GPU even when it’s working properly.

Have the same problem with nothing showing on the high setting for “unmix song”. I set the processing to cpu only as someone else mentioned and then the high setting works. It won’t work properly set to auto.

same here. only on high quality and when using the GPU (RX 7800 XT) . seems like this only happens with AMD graphics

Same here with flac when set to high, tried a 96k 32bit float file and same result on high, not tried the other settings. It does however take 5 minutes or so to come up with the empty layers so doing something. I have an AMD Radeon 5500xt.

I have a 2017 iMac Pro with AMD Graphics, 64 Gig Ram, MacOs 15.5, Cubase 14.

Spectralayers 12 really gets twisted up when I try to use the transcription functions in its various modes, ARA, Stand Alone, High, low.

It variously hangs the UI in MacOS, system becomes unresponsive , etc. If it produces any transcript at all it’s garbage.

Same results here. :crying_cat:

AMD Ryzen 5 3600XT
AMD Radeon RX 7600

1 Like

It would be helpful for the programmers if you could list your graphics card.

It seems that the AMD family isn’t the only one experiencing issues with these settings. There are other threads describing the same problems.

Until then …

Greetings, Jonas

I successfully used Unmix Song with AI Processing Device set to CPU.

AMD Ryzen 5 3600XT
AMD Radeon RX 7600

Hi @paulboco,
In this thread it is stated that despite the selected AI graphics card settings only the CPU is used

Maybe you will find some of your findings there or can provide further information.

1 Like

Hi @Jonas2021 . When unmixing Brass instruments, my RX 9070 XT is used, averaging about 35% but spiking into the 40s. Unmix time is about 8 seconds. Quality = High. Unmix Brass and sax first is selected. Same track with the CPU selected takes around 30 seconds and CPU usage is about 50% - GPU is 0-1% -same settings. Soundwise I can’t tell the difference between the two.

AI Processing = Auto

Layers are processed as expected - no missing audio. A mix of CPU and GPU is used, but mainly the CPU

EDIT: This is what it looks like in SL11 Pro

2nd Edit (more info):

Same settings but choosing Vocals only. AI Processing forced to GPU instead of Auto. The GPU is used, but results are empty.

2 Likes

Hi @Tanant,
Very good reporting/diligence.

Steinberg’s statement graphics cards for the AMD family:

Hope we get a timely update for these issues.

Have a nice one.

:wink:

DIRE issues, incredibly slow and poor results here.

30+ mins to BADLY unmix a clean/modern piano house track. It split the piano in 2 (Piano and Other) and a solo’d percussion sound was mixed in with the “other” piano too.

MAC / M4 Pro
64GB unified RAM
14 core CPU (10p/4e)
20 core GPU

I hear Robin is waiting on some help from Apple because Steinberg cannot get the MAC version to work properly (30+ mins to unmix is absurd). If that is the case, then you should not be releasing a paid product update on the Mac that has known issues. The complete lack of any feature updates free of charge since version 7 makes SL worse than Waves. I have paid for six yearly updates since version 6 and I am honestly disappointed. This is the LAST year for me and if the issues aren’t resolved in 14 days, I will be seeking a refund.

1 Like

Same problems as I before mentioned:

GPU: AMD Radeon RX 6700 XT, Driver 25.6.1
CPU: AMD Ryzen 9 3900XT
32GB RAM, Win 11 24 H 2, Build 26100.4484

In “CPU” only it works VERY SLOW, but the results in splitting duett in a male and a female voice are very disappointing ( in fact, they aren´r really splitted ).

For the ( German ) Steinberg team: Entmischt mal “Hallo kleine Maus” von Saso Avsenik…

Hi! of course, I am not a member of the beta testers team for version 12. My question is, is it possible to implement the ability to use graphics cores on Apple processors with a (Max. Ultra)prefix in future updates? The division of a track into stems takes a long time and the graphics cores are practically idle). Logic Pro separate the track into stems much faster.
Thenks!
Nikolay

Done!

I am deeply impressed by what SL can do!!!
Where will we be in 5 or 10 years fron now?

I can confirm the issue many others have. In the Options I did set the AI-calculations to be done on my GPU (NVIDIA GeForce RTX 2060) but everything is being calculated on the CPU (100%Workload).
The NVIDIA stays cool at 0% workload.
Sometimes the previews are stopped because of realtime workload.

The second issue wich is even more annoying is that in the modules the slider for the amount of reduction applied does not have any influence. What ever the slider is set the module is always applied to 100%.
No fine- tuning of modules is possible.
In these two aspects the previous Version 11 of Spectralayers worked flawlessly.

Specs

ASUS ProArt z690
Intel 12th Gen Core i7
NVIDIA GeForce RTX 2060
32 GB RAM
Windows 11 Pro
Nuendo 14.0.32
Spectrlayers Pro



12.0.0

Me too (seeking a refund) unless these problems are addressed.

@Sunnyman

And the male and female voices are on two different layers ? completely seperated ?

Yes, easy-peasy.

  1. Unmix Song

  2. Merge non-vocal stuff to make it look nice and rename this to Non-Vocals


  3. Take the mixed vocals

  4. Unmix Chorus

  5. Done - pretty decent, I think.

There are some artifacts, but it’s “okay” for this fast work.

My trouble with SL are on a different side: I need the VST-interface smooth realtime (like in Cubase or Wabelab) and not this stutter step crab… also the signal generator should be 100% reliable and the interface itself without any glitches. The last part of the interface might be a developers nightmare, though.
One thing in the interface bother me since I bought SL. When selecting "m"ono or "s"olo, one is not allowed to click and move too fast. Seems like there is something with the onclick-event chain.


I can not demonstrate this dynamicaly here.
There is much intern going on when clicking an element. mouse down, mouse up, mouse movement, etc.
The interface needs refinement for sure. A nightmare for the developer… I know.
Anyway, “it’s working”, so why bother? There are far more interesting things to do, than doing creepy coding stuff in mouse-events, right? RIGHT!

Also when “scrolling” is selecting, it should scroll when playing and not stop scrolling with the cursor running outside of the window.
This happens when resizing the spectrum or doing other stuff.

These kind of glitches in the gui are a strong sign for quick and dirty coding (which is not a bad thing in itself, but in a final release this should be cleaned up)… I know for myself.

I am tired of reporting all those bugs, because this is in the responsibility of Steinberg and its software testers. This is not the job of the customer! That’s not, what I paid for.

If Steinberg wants a good and thoroughly testing, sure, they can hire me. 200€/hour and I find all the stuff you may have missed. Promised, but you wont like it!