Spectralayers 10 Pro

[Guy Rowland]

Another day, another new suite of AI audio tricks. Spectralyers 10 has several innovations - separating overlapping voices, separating drum kit elements, separating piano and guitar and more trickery.



This is the first release that really interests me... there's a crossgrade price from RX.

https://www.steinberg.net/spectralayers/new-features/

[wst3]

call it old age, or stuck in the past...

I think I've tried every version of Spectral Layers since the beginning (can you say Sony/Sound Forge). It is clever tech, and in the right hands it is probably very useful. I've just never gotten the hang of it. I'm kinda stuck in the good old 2D, amplitude vs time view.

Which is odd, since I've no problem using a variety of views in my measurement work. Many are even further from old faithful.

I think part of the problem is I still make most of my edits by ear, for audio I've never been completely comfortable editing visually.

Once upon a time the visual representations were not all that precise. Now they are, but I like to hear what I'm doing.

If someone has a great explanation of uses for Spectral Layers I am all ears! (sorry, couldn't resist)

[Guy Rowland]

I think I’ve fully crossed over to visual editing, Bill - a decade with RX has done it to me. To the point where sometimes I zip a file across from PT, look at the spectrogram, edit, send it back without bothering to listen in RX at all. Then I play it in situ, and it’s right. I know what sounds look like I guess.

I actually get a little thrill of excitement at the thought of being able to separate that visual mess into layers, especially when it comes to overlapping voices. I know the pain of manually trying to discern one voice over another, and clumsily editing, which is never very good. So if you want an example use, that’s it. Previously impossible, essentially (if it delivers in practice - and early word seems to be good).

Spectralayers has always sort of promised things like this are possible, but until now the reality seems to have fallen far short of the potential. I do think this might be the version to have done it. To me its another example of iZotope dropping the ball. You have to go back years to find a really groundbreaking and useful feature.

[thesteelydane]

I think I’ve fully crossed over to visual editing, Bill - a decade with RX has done it to me. To the point where sometimes I zip a file across from PT, look at the spectrogram, edit, send it back without bothering to listen in RX at all. Then I play it in situ, and it’s right. I know what sounds look like I guess.

I actually get a little thrill of excitement at the thought of being able to separate that visual mess into layers, especially when it comes to overlapping voices. I know the pain of manually trying to discern one voice over another, and clumsily editing, which is never very good. So if you want an example use, that’s it. Previously impossible, essentially (if it delivers in practice - and early word seems to be good).

Spectralayers has always sort of promised things like this are possible, but until now the reality seems to have fallen far short of the potential. I do think this might be the version to have done it. To me its another example of iZotope dropping the ball. You have to go back years to find a really groundbreaking and useful feature.

I'm the same as you, I consider myself an RX Ninja, but the lack of something truly new in a the last couple of versions is worrying. The most infuriating thing for me is how to gets slower and slower the longer into a session it gets. I assume it has to do with the undo list, but there's no way to disable it, so I have to save and re-open my session every 20 min, or things like the spectral repair tools get's excruciatingly slow. Working on a 10 mic positions project now, which the composite view is made for, but after just 20 edits or so it crawls to a halt. Maybe it's just my system (specced up Intel Mac mini 2020)

[Guy Rowland]

FYI there's a free 30 day trial up now - https://www.steinberg.net/spectralayers/trial/

A music test was pretty darn impressive - the separation between guitar and piano is quite incredible at times. In general I think it is superior to RX9 (I don't have RX10 but I believe it's the same), though swishyness is still frequently present.

Much more disappointing however was separation between multiple voices. I gave it a pretty easy first try between a male and female speaker, but it failed to to differentiate at all, marking them as a single speaker.

It'll be playing in more depth in the coming weeks, but my urge to get it sooner rather than later has diminished - some way off from that particular miracle coming true it seems.