I recently purchased purchased SpectraLayers Pro to use its de-bleed and unmix multiple voices modules.
I’ve run into an issue where when I export my finished audio (happens for both export layers or export mixdown) the resulting file has a bunch of 0 db, 1 sample long clicks/clips at random intervals during moments of silence.
Original Project in Yellow, Export-only Clicks in Red
The issue also occurred in the demo of SpectraLayers 11 Pro.
I believe this may be a quantization error resulting from SpectraLayers projects being in 32-bit Float format and then my exporting in 16-bit Int format to save space (the source audio is 16 bit to begin with), which would explain why the clicks aren’t in the project itself but are in the exports. As evidence, if I choose to export a wave file in 32-bit Float format, the clicks are not there (or are so quiet as to be barely noticeable).
Naturally, the obvious solution would be to use dithering when exporting to remove these quantization errors. I’ve experimented with all the different dithering methods available for wav and flac exports, and none of them have removed the clicks.
Considering this software costs $350 dollars, I am very disappointed that exports are broken like this, especially considering this has never happened to me in other audio editing programs. I’ve scoured Google as well as these forums and can’t find any record of anyone else encountering this issue, which is all the more frustrating.
Long story short, does anyone have any suggestions or solutions for this issue? Thanks.
My Specs:
SpectraLayers Pro 12.0.0 Build 421
Windows 11 Pro 24H2 OS build 26100.5351
AMD Ryzen 9 7590X
RTX 3080 12.0 GB
64.0 GB of RAM
Hi!
Could you upload the original audio file?
Yes please, provide a 32bit float wav or a .slp project so I can try to repro locally the issue when exporting to 16 bit wav.
Same, will test this end for you too
Thanks for reaching out, everyone.
The file size limit is 4MB, so I trimmed it down to a sample that exhibits clicks on my end when downconverted to 16-bit. The file is in .slp format in a .zip archive.
I’ve included the unedited audio, the de-bleed profile I’m using with the de-bleed module, and the edited audio with the de-bleed module applied.
mez1994 Sample Project.zip (1.8 MB)
A little more information: I’ve played around, and it seems as though the de-bleed module is responsible for the clicks appearing in exported audio. If I import a 16-bit file and then immediately downmix it without performing any edits—or perform simple edits with the eraser tool—the clicks do not appear.
Hi mez1994,
just to confirm I can reproduce what you are experiencing: 3 clics on the “Michael - Debled”. It happens with the 16 bits and 24bits export, not with the 32bits export.
Also confirming on a Windows 11 system.
Here my findings.
Seems like the “clicks” come from “missing” parts in the debleed layer. At least it looks like that there is no data at all, instead of just zeros (minus infinity).
In the pictures given I exported as 24 bit dithered and there is no difference to non-dithered, the clicks are still there.
Interesting: Before export I did try to merge the debleed layer with another layer (complete silence and noise) to fill the gaps and get a better result, but without success. The “missing” parts seemingly grew even bigger.
More interesting: In Composite Layer View the “clicks” do not appear in the spectrogram, the composite waveform remains broken, though.
Only when soloing the exported file layer I will see the “clicks”.
Even more interesting: Looks like the 32 bit float export is not completely unaffected either. When opening the 32 float file in WaveLab, at some zoom levels I can clearly see the three large “clicks”, spiking at about +30dB according to WaveLab’s auto-zoom function. More confusing is, that the waveform is displayed simultaneously as positive as well as negative, which should not be possible (?).
Zooming in to sample-level or doing a Global Analysis on the faulty sections reveals just zero (minus infinity).
For this
Did you check this with Audacity also? I would recommend doing so.
When it comes to numerical accuracy I do not have full trust in WL and SL. This is very sad, but that’s the way it is.
Steinberg should improve its software quality before delivering more and more features. The software seems somewhat lost by trying to be a jack of all trades.
Edit afterwards:
It’s “ok”
This can in principle be constructed in Audacity.
Sincerely apology to Steinberg on this case here.
But in general, my statement about software quality stands, though.
(I am using WL, SL and Cubase almost everyday, so the complain could be taken as a compliment. I would not complain, if I wouldn’t care. You are doing a hell of good job, All the complaining is certainly some kind of a luxury problem. But as long as I have the luxury of complaining,…)
I could also reproduce it!
It’s technically a rectangle signal and “okay”.
Thanks, after analyzing the issue it seems to be an issue with the DeBleed module rather than export options. In that case, it created NaN samples (samples that don’t have values), therefore the export could not properly quantize that. It’ll be fixed in the upcoming patch (you would have to apply DeBleed again on your original audio though).
2 Likes
Robin, thanks for looking into my issue. I look forward to the patch when it comes out.
