Apple is still working on improving support for that… We’ll see what happens later that year.
2 Likes
sigh. Thanks anyway.
Robin,
Perhaps offer a few hints of what is to come to whet our appetite.
As per BJ Dobbs I also can hardly wait!
1 Like
Yes, just a few crumbs of info to keep the pangs away. Any interesting AI developments? Will you finally exceed RX’s feature set e. g. separating multi-vocal parts? Are we not intrigued?
Out of curiosity, how do you rate it? In the meantime I find myself not using RX for seperation.
2 Likes
More unmixing, better unmixing. And so much more in lots of different areas. I guess I can’t tease more at this point, but it’s getting closer, and it’ll be worth the wait - probably the biggest SL release to date 
9 Likes
Generally, I consider it a breakthrough product and essential to my workflow. It does what I dreamed and hoped software could do many years ago. The fact that it’s part of the Steinberg family is a big plus. Using SL to its maximum potential feels like an art form on its own. At a deep level, Spectralayers can be a formidable creative tool.
I saved many one-off recordings on cassette and reel which were great takes but with sonic flaws that could not be fixed by conventional mixing. Spectralayers makes me feel like George Lucas digitizing and restoring early Star Wars. It opens up so many possibilities, it almost makes me breathless.
Apart from music, I edit video. SL is amazing in a film environment but not quite up to RX. My hope is for SL to achieve higher repair capability and voice/artifact separation.
Also, Robin’s stellar attentiveness and care towards SL users is a major bonus.
2 Likes
You asked “Will you [Robin] finally exceed RX’s feature set e. g. separating multi-vocal parts?”
I meant, “how do you rate RX” … because I don’t think RX can hold a candle to SL10.
(i.e. RX doesn’t have any ability to seperate multi-vocal parts, as far as I know).
I’ve heard RX pro has better separation, isolation and repair than SL. It is widely used in film to remove unwanted artifacts e.g. a perfect take but the gaffer dropped his keys from a catwalk. I’m told it’s useful for separating speech with different timbres. I haven’t tried RX10 yet but RX11 is coming soon and I may demo it. I’ve used earlier versions like RX8 with good results at the time.
So far, Spectralayers does the trick for me and I’m hoping it will surpass RX. SL is better at spectral editing which suits music better and it does a great job with film sound tracks.
New versions of each will appear around the same time later this year. I have a reference clip from a doc film that was scrapped largely because the sound guy didn’t know what he was doing. It’s a minute long clip of a narrator with a deep baritone voice like James Earl Jones speaking beside a diesel bus with engine running. So far, nothing has been able to perfectly separate them. I’m eager to learn whether RX or SL will have AI models that are up to the challenge.
Hope I answered you satisfactorily.
1 Like
On voice cleaning and separation, you can find a comparison between SL10 and RX10 here:
https://divideconcept.github.io/Restoration-Comparison/
(AFAIK RX can’t separate multiple voices)
1 Like
According to many who have tried both that’s no longer really true.
Combining SL with various Acon Digital plugins will fulfill most expectations.
1 Like
I’ve used RX versions 6 through 9 and have tried the demo of RX 10, and I find SL10 produces far better results for manual cleaning of dialog. In my opinion, RX 10 does not seperate individual voices, although I’m sure with enough time and patience with the spectral editor, one could achieve some result.
RX, on the other hand, can be faster to use in certain workflows (its plug-ins can be used in other applications, for example) but only if the demands are not too high. I find the “traditional” restoration tools in RX (declick, dehum, spectral denoise) especially useful and quick and easy to use in most restoration jobs.
In the same vein, the new Repair Assistant in RX could be great for some use cases, however I personally don’t have a use for it and would prefer to know, and have control over what’s going on at a granular level.
So, different tools, and I don’t think it’s necessarily an “either/or”, it depends on what is best suited to what you need to do most. I use both and will be trying the demo of RX 11 when it’s released.
1 Like
Izotope underlying code is spleeter. Sl underlying code is demucs. Last I heard.
Neither works perfectly for all sources and some results depend on how a developer further tweaks underlying code.
I personally like demucs results compared to spleeter…on most of the somewhat-distorted-mono 1960-1965 projects I deal with.
I get fairly horrible drum demixing on all of them (bass drum, hh, toms, snare, rides, crash)…but again…the source material is mono audio…I have manual ways to effectively extract each drum once the overall song is demixed…and…I know that over time, automatic routines for separating drum pieces will get better.
Regardless, the demix tools brought to market in just the past four years ansolutely dwarf what I was using ten years ago (Prosoniq Isolate).
I think the key at the moment is to buy several programs that use competing underlying code (spleeter-demucs) and always test which works best, given varying source material.
1 Like
Just giving you a heads up, isolating voices is EXTREMELY HARD! (even with neural networks) . Even with all this A.I. craze, isolating voices is HARD and I wouldn’t expect that anytime soon.
Unmixing bus/sum tracks (like stems) of songs is much more easier to do because things like drums(transients) and bass is more easy to distinguish and these algorithms are using masking techniques. A human voice is literally like a human fingerprint (no 2 prints are the same) and the same technique that is used to separate stems won’t work on vocals because there’s way too many variations on a human voice. The only way I could see a way for an A.I. process to separate voices is the algorithm would literally have to change and be based on harmonics/overtones(the timbre of a voice print), the method to separate stems simply wont work on things like the human voice (there’s too many variables and the voice is much more complex than people realize).
1 Like
Most available code works fine for yanking a single voice off of a mixed track.
Maybe you mean that separating multi-voice harmonies is more difficult. Which certainly can be.
1 Like
Looking forward to SL 11. I like Spectralayers, and have stuck with Steinberg because of the consistency of releases. Plus how well Yamaha has done with them. With RX, I don’t use Izotope products personally anymore because of the current owners and how poorly things have gone with NI/PA.
1 Like
Better Noise Reduction with more parameters.
It’s not even as good as the noise reduction in the old RX8.
I’m using it mostly to remove static noise (typical “USB noise”) from external synths and RX can easily capture the noise profile. In SpectraLayers I have to boost the signal about 20dB and then such noise profile produces typical noise reduction artefacts on quieter signal, not to mention artefacts on reverb. RX doesn’t make any artifacts while completely removing the noise.
Statistics (LUFS, Peak etc) in Elements 
RX has it in the Standard version. Common, it’s a fundamental feature that even version One should have.
I want to quit RX but for now, it’s impossible. I hope that 11 will change that.
1 Like
I am with you on that. I have been an Izotope fan for years, but the last 3 years or so, their marketing practices, particularly their pricing by way of product bundles, has been incoherent, counterintuitive, and too expensive for the value I got out of it. I really would like to stay current with Ozone and RX (the rest I can take or leave), but they just don’t seem interested in taking care of long-term customers.
And it is worse under NI. Like Izotope, NI is a company that has been innovative. But they seem very committed to the proposition of pulling customers into their proprietary ecosystem. I’d much rather remain in the open space where we can pick tools from multiple suppliers and have some confidence they will work together gracefully.
For me, the upgrade pricing for RX11 is just not worth it, and I am one who usually upgrades my “go to” products within a few days of announcement. I wish Steinberg well in advancing Spectralayers. It is clear that RX has a quality difference at present when it comes to noise reduction and some cases of unmixing. But Spectralayers has a superior (for me, anyway) concept by naturally retaining all the individual layers of the unmix. (I’d like to see the same approach carried across to everything, including the noise reduction – i.e. I’d like to have the option to click a checkmark to include a noise payer in addition to the de-noised layer.)
Heck, if this comes soon enough that I can avoid buying RX11, maybe I will finally spring for a Wavelab license.
1 Like
I beg to differ… The Voice DeNoiser in SL10 is clearly superior to RX10, and so is its unmixing capacities. And while I can’t disclose much about SL11 at this point (even though it’s getting closer !), there’s an even bigger gap between SL11 and SL10 than there was between SL10 and SL9 
8 Likes
My impressions have been that I got better results from RX10 when doing general de-noising of a live musical recording. But they are close enough I don’t wish to debate the point. I look forward to the continued progress.
I would rather spend my money on SL than RX at this stage. I have found the de-mixing in SL has allowed me to accomplish some things that were not possible before. Speficically, there are some occasions where I want to do a good quality live recording, but the stage arrangement or performance schedule doesn’t allow me to set up individual mics on each instrument and voice. In the past, I would use a stereo mic for the main sound, and then maybe try to put a splittter on a vocal mic or 2, and if I was really fortunate, maybe get a mic on the kick drum and a DI off the bass.
Put in those cases, the main stereo mic usually was flooded with drums. With unmix, I can take the vocal completely out of the stereo mix (because I still try to have the vocal mics separately.) And most importantly, I can take the drums off to a layer I can control separately, leaving the rest of the band in the stereo mix.
It isn’t a studio quality result, but I have done this 5 or 6 times with surprisingly good results each time. It is reliable enough that I may opt for a simpler recording setup even when I didn’t face time or space constraints. For me, unmix is transformational, and SL is far more useful than RX in this regard. I do hope the maturing versions get a little better finding the bass more consistently.
1 Like