SL11 De-noiser

I was looking at this video below:
SpectraLayers 11: Noise Reduction and Song Unmix Testing While Searching for Alligators - YouTube

Don’t know the guy, but was searching for SL11 videos on Youtube, just to hear / learning from other poeple, how they were getting along with SL11.

In the video above, am I hearing right… when the noise reduction is ON, it also removes from this guys voice?! It’s like his voice is getting more thin.

For sure the noise is gone… but it’s also like the guys voice is getting hit, or am I wrong?! :slight_smile:

All the best,

When using Voice DeNoise with Background: Noise (Strong) SL11 is using a more aggressive AI to get rid of the noise, which might slightly lose some of the voice during the process, but my interpretation here is that the main audible difference here is not due to that, but to the absence of heavy noise around that makes it sound a bit different.

Specially when the audio comes from a compressed format, which is likely the case here : compression algorithms (such as mp3 and such) keep the most predominant frequencies and skip the quietest ones. So when recording voice in a very noisy environment, the audio compression algorithm might choose to keep more noise frequencies than voice frequencies… And when that’s cleaned up in post, you end up with a voice that sounds like it’s missing some frequencies, because they were replace by noise frequencies in the compressed file.

Not sure if that’s entirely clear, but hopefully you get what I mean :slight_smile:


Thanks Robin for the explanation :slight_smile:

Yes, it makes good sense, that heavy processing might take a little from the rest of the audio. And also, that it probably wouldn’t be necessary to use the strongest settings. Compressed audio couldn’t be a problem, as you say.

I might be a little direct on things, but I just want Spectralayers to be the best tool, in terms on song separation and restoration stuff. And I guess all info that can show differences between different tools, could help in bringing SL up on top of everything else there is.

I think SL’s closest competition is RX11. But I do like SL’s GUI more, I think it’s easier to work with. And also the AI on song separation, is way ahead of RX11 :slight_smile:

When everything comes to the conclusion, there is always space for improvements. Right now, I guess the bugs fixing / improvements in the upcoming patch, is number 1 priority.


I think it’s important to understand that the quality of the source file can be decisive in determining the outcome of all noise reduction processes, not just AI-based ones.

For example, a 24-bit 96kHz recording will contain everything the microphone picks up. This gives the noise reduction process the maximum amount of information to work with.

As soon as any audio data reduction processes are applied to that original recording, the total information content of that recording is reduced (this is why file size can be smaller) and that means subsequent noise reduction processes are also missing potentially vital information.

Phone recordings or MP3s are not giving the software a fair chance.

If only things were that simple.
Often we have to use several tool to get the results we are after.
For a recent documentary we ended up exporting several passes from FCP and opened them in SLP9. Then using the excellent spectral selection tools we manually refined and fine tuned the voice print, and in the end using the same tools extracted and merged the ambience back. We ended up using roughly 20 layers in SLP9.
It took us 2 days just to fix a few seconds of dialogue.
So AI tools are great, but the real hardwork start after…
We did a fair and honest comparison here.

I admit that this was a hopeless scene, and there are no ADR in a Documentary film nor did we find matching words that could have worked as a patch. Today we used SL11 and was able to archive the same results in fewer steps and layers in less then 20 min. (And a detailed tutorial is on the way)
Was it better, no it was as good as the source
You cannot make it better than the source but you can definitely improve on the intelligibility.

The AI tools are getting smarter day by day, my only regret is that thay are not trained sufficiently on Indian languages and so are not so efficient with it. The Hindi transcription is not as good as the English despite Hindi film industry being the largest in relation to volume output.