SpectraLayers 10 vs RX vs RipX?

[Caine123]

Yep, SL10 is the one to beat right now.

Thinking of doing the 'competitive crossgrade', how does that work ? Does anybody know what 'proof of eligibility' is needed ?

yes you need to prove that you got one of the following
Adobe Audition (Single App and Creative Cloud subscriptions)
iZotope RX 5 and later (Standard / Advanced)

just upload/send them an invoice or something to prove you got it.

Ok thanks. I own RX7, but don't know if i have any invoice or something. Guess i just see what Steinberg asks for.

try to send them an invoice if you find it, paypal transaction maybe of this if you got or just proof that you got the license in your Izotope account (picture or something i guess) and it should e enough i guess .

hope it works out, i think they might be happy to sell a license and get another Spectralayers user so i dont think they make problems.

[noiseboyuk]

If you want to give it a try, SpectraLayers 10 trial was just released: https://www.steinberg.net/spectralayers/trial/

Thanks for the heads up. The unmixing is very good, definitely better than RX at this point, though there's still some swishyness. However the separating voices I found plain didn't work - fed it with male and female voices - low hanging fruit I'd have thought - and it couldn't differentiate.

[Robin Lobel]

However the separating voices I found plain didn't work - fed it with male and female voices - low hanging fruit I'd have thought - and it couldn't differentiate.

If your goal is to separate overlapping voices, you can use Process > Voice Denoiser > Background: Other Voice, this one should work in all cases (that's all the devoice examples here: https://divideconcept.github.io/Restora ... n/#devoice )
For speaker identification in a dialog, there will be improvement to the algorithm in further patches.

[noiseboyuk]

However the separating voices I found plain didn't work - fed it with male and female voices - low hanging fruit I'd have thought - and it couldn't differentiate.

If your goal is to separate overlapping voices, you can use Process > Voice Denoiser > Background: Other Voice, this one should work in all cases (that's all the devoice examples here: https://divideconcept.github.io/Restora ... n/#devoice )
For speaker identification in a dialog, there will be improvement to the algorithm in further patches.

Thanks Robin - you're on the Spectralayers team?

I just tried that method and it was a little better, but typically would make one voice sound "crunchy" for want of a better description and leaving the other alone. But even that was variable - in some cases whole words or sections of words would be missed.

It all feels like technology that is getting there, but not yet reliably delivering. Love the direction of travel though, and it's really successful in other areas.

[Robin Lobel]

Thanks Robin - you're on the Spectralayers team?

I'm its creator and only developer since version 1

[noiseboyuk]

Thanks Robin - you're on the Spectralayers team?

I'm its creator and only developer since version 1

Why hello! That's amazing - the ONLY developer. Staggering how you're beating iZotope and still innovating. All power to you!

[kelvyn]

My initial thoughts after demoing on two tracks are of feeling a little underwhelmed by all the candidates.
I was hoping spectral layers would be able to differentiate between a guitar - in a mix - and some noise from cymbals.
Is there a way to separate the audio file again?

[dionenoid]

Thanks Robin - you're on the Spectralayers team?

I'm its creator and only developer since version 1

In that case, if i may ask : Are there any ways (or plans) to unmix stems which don't fall in the standard categories ? Modern music often doesn't make use of standard piano or guitar but unmixing often mistakes other sounds for this. Especially with electronic music, where synth sounds are very often mistaken for vocals, or kicks partly mistaken for bass sounds, or attacks from bass/synth sounds mistaken for drums.

For example when trying to unmix house or techno music, i see that a kick is often part in the bass category, part in the drums category, and part in the 'other' category, with some even parts in guitar or piano stems. When doing the separation thing, it also mistakes this, not really separating well into tonal, transient and noise stems.

RX and Acoustica let you adjust the amounts, or the intensity, but i couldn't find something like this in SL10 (neither in RipX) so i guess this is due to the algo used (Spleeter or Demucs) ? Or are there any settings i can change ?

[Robin Lobel]

My initial thoughts after demoing on two tracks are of feeling a little underwhelmed by all the candidates.
I was hoping spectral layers would be able to differentiate between a guitar - in a mix - and some noise from cymbals.
Is there a way to separate the audio file again?

It usually has no much trouble separating drums (and vocals) from the rest. But if you run into a specific example where the separation didn't provide satisfaction, then that's where it's time to use all the spectral tools SpectraLayers provides on the left side of the UI.
Since its version first version SpectraLayers has always been about giving the user surgical control over the spectrogram to separate and rebalance elements, to overcome what automated processes can't do.
For instance on this page I provide a couple examples of manual rebalancing after AI unmixing: https://divideconcept.github.io/Unmix-Comparison/

[Robin Lobel]

In that case, if i may ask : Are there any ways (or plans) to unmix stems which don't fall in the standard categories ? Modern music often doesn't make use of standard piano or guitar but unmixing often mistakes other sounds for this. Especially with electronic music, where synth sounds are very often mistaken for vocals, or kicks partly mistaken for bass sounds, or attacks from bass/synth sounds mistaken for drums.

For example when trying to unmix house or techno music, i see that a kick is often part in the bass category, part in the drums category, and part in the 'other' category, with some even parts in guitar or piano stems. When doing the separation thing, it also mistakes this, not really separating well into tonal, transient and noise stems.

RX and Acoustica let you adjust the amounts, or the intensity, but i couldn't find something like this in SL10 (neither in RipX) so i guess this is due to the algo used (Spleeter or Demucs) ? Or are there any settings i can change ?

Further fine-tuning improvements and options are coming with patch 2.
But we'll eventually reach a point where it's not gonna be possible to train an AI on every possible instrument specifically - there are just too much, and a lot sounding too similar to each other for this to be accurately possible, not to mention the amount of training data that would be required.

However to overcome this I have some thoughts for SpectraLayers 11 - but can't reveal much at this point.