Ooohhh tomorrow 1pm BST , i’m looking forward to this , a nice Bday prezzy for myself 
So what happens when you select a frequency (using the magic wand tool) and go to the menu and select the processes tab and select amplitude and then choose gain?
What I do is improvise. A tip that is useful is using the magic wand tool. Select the magic wand tool and then set the tolerance to 3.0 db and maximum width to 4.00 s, I find that is where the sweet spot is and most times it picks up everything from every audio. Those settings also (surprisingly) recognizes large transients and large partials and you can improvise by holding shift (as you’re selecting). When you hold shift, the magic wand tool really is powerful.
When you say merge selections, what do you mean exactly? You can already merge selections by selecting Add To Selection. If you’re talking about (or you mean) a straightforward merge then I already suggested this, but first a few issues has to be fixed. One of the issues is that selections has to be redefined within Spectralayers (as a state of its own), as of right now if you open any audio file and start making selections and then save that project file (without actually saving the selections) then when you recall that project file all selections will be lost. That would have to be addressed first before you can do a straight merge
This is something that I wanted to suggest but I didn’t want to overburden the developer. I noticed on my surface pro that there is a distinct lag when drawing selections. Then on-top-of-that, when you make many drawing selections(400+ selections) the cpu/gpu starts choking and throttling and stuttering and glitching(buffering issues). This is something that I brought to the developers attention but honestly I feel like something like optimizing Spectralayers is beneath him and that job should be allocated to someone else. The lagging issues(using a tablet) and optimization should be the responsibility of Steinberg and I feel it is Steinberg’s responsibility for picking up the slack and getting Spectralayers more optimized (not the main developer’s responsibility).
Use the Magic wand tool. The sweet spot is setting the tolerance to 3.0 db and setting the maximum width to 4.00 s, then use the shift shortcut.
Yeah, I noticec that as well. It’s a bit jumpy (just like the playhead).
2 Likes
Talking about merge in terms of when you have selections you need to repeat. Select similar doesn’t really work very well for some situations. I’d basically just like to draw my selection on the pattern, then ctrl+drag to duplicate it and move to the next repeating pattern. In terms of the menu, you can only select a single saved selection, it would be nice to have checkboxes next to all selections so you can place them in one hit rather than many clicks backwards and forwards to keep adding to the selection.
Magic Wand isn’t accurate enough for things like a siren or very tightly recorded notes, it’ll always stray as it hits other sounds at the same frequency, so a repeating pattern and very narrow frequency selection with simply selection shapes would be more accurate and faster.
Select similar works perfectly for all solutions. I’ve never ran into a situation where it never worked. Can you post an example of a situation where select similar didn’t work. Can you post a project file (where you have the selections saved for select similar) so I can get an insight into what you are talking about when you say “it doesn’t work well in some situations”.
I did a tutorial (it was actually my very first video) demonstrating how to exactly do this but had to remove because of the music (newsflash: royalty free music isnt exactly free and I learned that the hard way).
You can make a selection and then save selection and then drag that selection to the part where you want. You can look up in the top right corner to see what the frequency is and amplitude number is. You can also turn on cursor crosshair and cursor coordinates to precisely drag-n-drop to where you want to drop. It kinda sorta snaps, just look for the last digits(and mark it down in your memory) and drag to the same coordinates(frequently wise).
Yes! I suggested something similar to this to the main developer. In order to “kill 2 birds with one stone” I suggested a feature for the option to lock the axis all 4 ways(bidirectional and horizontally and vertical and diagonally) as you’re dragging and dropping. That way you can still resize the selection (while it is still locked on its axis all 4 ways). The best way I can summarize (so the developer can get a picture of what to implement) is to think of lock points. This video is a great way to illustrate what I mean by lock points. If the main developer can somehow integrate region markers as lock points and you’re able to resize selections(and even multiple selections) without interfering with(or moving past) the lock points, this would literally “kill 2 birds with one stone”. As of right now, you can resize the selections but the resizing is done in a scalable linear fashion, you cannot resize the selections while having the option to lock the axis. Meaning, if you wanted to scale or resize any selection but choose to add points/markers (where it locks the axis and does not scale or resize) then you cannot currently do that.
Would like to prove you wrong. Can you post an example(project file) of where the magic wand tool failed you(in particular the example with the sirens you keep referring to), so I can take a look and maybe do a mini tutorial to demonstrate how powerful the magic wand tool is?
1 Like
Not going to start posting examples, I simply don’t have time, sorry. I have been using SpectraLayers since Magix days though so nearly a decade and those are the shortcomings I experience every day. Having shape tools like straight lines, bezier curves and basic shapes is faster and more accurate than a magic wand, lasso and brush because some sounds do not need a threshold, you know where they exist, you just need a fast way to select using the touchscreen and a pen copy>drag through the song to repeat it ratherthan this game of addition all the time from saved selections. Simply ctrl+click>drag doesn’t work and often you need to select things from an empty layer to easily move selections before comitting to an actual cut/copy. Select similar is good at many things, but it is still not accurate-enough for some situations such as where a sound subtely changes through several minutes even though it’s the same repeating pattern at the same frequency. Yes you lower the threshold of the similarity, but that is then increasing the selection window which you don’t always want, so then you end up manually editing anyway because its more accurate but frustrating/boring work.
I’ll post some examples next time I experience the select similar not working, I generally never find it works exactly the same as manually selecting similar which is more accurate and gives better results.
No it’s not. The magic wand tool can pretty much select any shape (it even registers and picks up noise elements, but like I said you have to adjust the settings).
You can manually move the selections around with the pen. Click on the transform tool and then select transform selections.
You can also continuously select selections and continously keep selecting without adding to new selections or manually saving the selections?
Yes please do. Also can you post examples of the magic wand tool not selecting what you want selected.
1 Like
Hopefully one day SpectraLayers Unmix Stems will be up to par with Ultimate Vocal Remover (which is free). It is the one with which I have obtained the best results (although quite slow). After watching the video presentation of version 10 I still miss in SpectraLayers being able to separate the main voice from the music but keeping the choruses (function that Ultimate Vocal Remover has, in karaoke mode). Maybe for the next version?
1 Like
You mean the reverb or background vocals? If you mean reverb then you can de-reverb within Spectralayers. If you mean for background vocals for spectralayers then you can use the harmonies/overtones tool. Also there is one other application that can sort of do this but personally I believe the approach is wrong (it also fails at most audio material I test with). Basically what they’re doing is building libraries/datasets for every scenario(strings, guitars, cellos, saxophones, lead vocals, background vocals, filler vocals, synths, distorted synths, etc) which is not only inefficient but is kinda pointless. The best way to solve that problem is to build libraries/datasets around sources, that way you can unmix a whole 50 piece orchestra and separate all sources(whether its harmonic vocals singing in unison of triplets or quartets).
Bit underwhelmed to be honest , going to wait and see what these separations are like because for me SL9 was worse than SL8
For us in post it looks like a very good update. For one if the de-reverb algorithm has improved that’s great. I use Acon for that all the time, and the more I can keep it within Nuendo / Spectralayers the better.
Another potentially great feature is the ability to transcribe dialog. I can see myself change my dialog track layout a bit to accommodate and then run the transcription and use that to find alternative versions of words as I do my dialog editing. If it’s done well it could be fantastic.
Between just the two of those I think it’s an easy buy for me.
Ok So Spectralayers 10 is out and does most all that was requested here (and our expert in unmixing: “Unmixing” was going anyway, many times out of careful manually selecting and tuning SL settings per tool), only now SL10 does it by improved menu options.
For me, the quest is complete and done!
MASSIVE Thanks Robin Lobel, after a bit more than 10 years from Spectralayers release, it now lives by its original promises.
This is the merge of communal efforts (various tools & AI) with a very well focused and able developer, Robin. It couldn’t happen other way and it did in quite fast timing, considering this achievement is of historic proportions: a revolutionary new audio tool being taken to the utmost professional usage level.
A few of us users and even less engineers and academics had been wanting, on the search of these abilities since several decades and now it is here in such a mature form.
@equisonus @Highly-Controversial SL10 is not Spleeter based. See the demo in vocal extraction quality SL9 vs SL10 in the World Premiere video. SL10 has state of the art AI for stems unmixing (and state of the art AI for drums unmixing, voice denoising, voice dereverb and multiple voices unmixing as well). It really is a tremendous increase in quality across the board.
7 Likes
HI Robin , i have watched it , it seems to sound great but i also need to compare it to SL8 which sounds better with least artefacts to my ears , i love 9 for the work flow but the separation is a lower quality than 9 , i will certainly be checking the real world workflow out
SL10 looks like an excellent upgrade - thanks for all the hard work.
1 Like
So … when is the release date?
Izotope is actually based off of a customized version of demucs.
June 22nd.
If Robin says his SL isn’t using Spleeter then i’m pretty sure it won’t be using Spleeter
To be clear : SL7/8/9 were using Spleeter, but SL10 isn’t, it’s a new AI.
However iZotope RX is still based on Spleeter (or a derivate of it).
4 Likes
Okay, I may be wrong. I was trying to look for the article where facebook confirmed that izotope is using their algorithms but I couldn’t find it. I do remember reading something about izotope using facebook’s algorithms(demucs) years ago but because it was so long ago I can’t recall what it was.
Even though I may be wrong, dont believe everything from Google. Google doesn’t have all the answers. Maybe they(Izotope) did end up licensing from Deezer for research purposes but I dont believe their algorithms were based off of spleeter. Another thing to keep in mind is that company’s like iZotope are in the market to compete against other competitors, so they’ll purposefully throw out misinformation to deter others from learning their trade secrets(for example claim their using one algorithm when theyre using another). However I could very well be wrong and I apologize if I am.