I use SL10Pro, now we got 10.0.10 with GPU support - check this :) I also use Acoustica Premium and RX9. I prefer SL10 for unmixing, but not impressed about de-noising in this tool.
@@GadesChannel you should give de-noising another try, it's probably the best of all 3 in that area as well. Can't link here, but I've set up a page comparing SL10 vs RX vs Acoustica. google: divideconcept spectralayers restoration comparison
Welcome to Warsaw Paul Land. Hey man. Hope you are doing well. Watching this now and have to say i regret buying teh Acon Remix plugin. I does nothing my Rx cant do. Only its handy sometimes to have it in the daw as a plugin but in reality i never used it there.
I would have liked to hear just the rendered separated parts played back together, like all the Spectral Layers pieces played together to see if it adds back up to the original. Fantastic job comparing them!
When listening to the drums of the SPECTRALAYERS, you can hardly hear the shakers, although the RX has them. But in terms of sound quality, SPECTRALAYERS surpassed everyone in everything.
@@FoliaSound Okay. Bad to hear. RipX primarily aims at DJ's and producers. Stems creation and sound(s) replacement and arrangement. Where I think RX and SL focuses on sound designers and sound restoration. BTW: I am not linked to RipX.
aaand... Dom Sigalas showed also that the extracted stems played together and tested nulling against the unmixed original track, that there is a perfect null with spectral layers! Same I see with own tests and that is simply awesome. I also tested this with HitnMix RipX but does not null.
Frankly - why would I need it? I need it to extract proper steams for further use. What's the advantage of the stems summing up to a null? What's the practical advantage of this? Is it good at all, because it means, that same frequencies are split and shared throughout different instruments? Doesn't it make them a bit hollow and empty, a bit like mp3 encoding? Serious question!
@@FoliaSound Dom Sigalas reffered to Mastering to see an opporunity there using it. Or kind of remixing, Dom put a delay for example on the voice, which wasn't there before and it worked perfectly. Also reverb, in general, is used in mastering sometimes, so if you can go a little more surgical, why not? Also when there is only a final mix left and the mix session or stemms aren't available anymore, nice to have a null. I for myself had some live recordings (single microphone, voice plus one instrument) where there was some room noise and talking and stuff and which I had then easily on (a) layer(s) which i could simply edit and/or delete, quick delivery was needed. So you can see it also in context of a restoration use case, if you will. Afterwards I was able to render it out merged back again. And when the extracted stems are in such a quality that they null gainst the original, bam! Same story with RipX, but I had artefacts being introduced in some cases. And comparing RipX against SL10 (which you hadn't in your test, I read about the reason), RipX would not have kicked SL10 from the throne, but quite often would have been on par.
@@FoliaSound In the extraction process, yes, frequencies are always shared between insturments, thats where a quality algorithm is needed, that cuts out precicely, but only what is really needed, a really surgical clean cut, that keeps part alive and musical, thought about that too, but... What I can say, at least, that I had that empty and hollow feeling a little with RipX, had the feeling like a little bit of liveliness was sucked out, got a little dry, dont have that impression with SL10 so far. But both were often, in my tests, quite on par, just nuances. I guess it also has to do with the quality of the track you want to split (mixing, format), obviouly. But what RipX and Sl10 both weren't able to accomplish was extracting the bells out of the Smashing Pumpkins song "Disarm". But anyways, SL10 makes best musical separation and leaves you with a null if necessary, whats not to like about it?
@@cpssounds3656 , I believe, that nulling when reverse separating is actually not a good thing and I can easily explain it to you. The fact, that separate instruments of your mix share frequencies over time doesn't mean that when you render them solo, they'll null when being added up, right? They create a mix because of all the routing, harmonic content and tonal and dynamic processing. When dissecting the track, using SL10, it's different - you separate from the sum of sound that you have, and then, in reverse, you simply sum the channels up. No routing nor any dynamic/tonal/harmonic work and character is recreated. Thus rendering a solo instrument from you master will always sound better and fuller that something that nulls. When it nulls, it means it lacks a lot of info about the instrument and its real content that has been masked in the master, but it's still there. In SL10 it's not masked - it's just not there. I'd realli like to see the elements of separation not nulling in SL, which would give me the hope, that this algorithm tried to retrieve something masked instead of sharing the master it had at the input. This is really not the best way to recreate the source instrument. Hope it's 100% clear on my side now!
@@FoliaSound Ah, I guess here is the culprit, that maybe brought in some confusion: "Afterwards I was able to render it out merged back again. And when the extracted stems are in such a quality that they null gainst the original, bam!" - "Afterwards", meaning after i removed room noise, talking and coughs and stuff, no fx, eqs or stuff - "renderign out merged back again", meaning only the cleaned stems (without the noise) rendered back into a single stereo track - "...they null against..." meaning, you have less artefacts in such processes, when the untreated stems alone null against the original, generally speaking - of course the extracted - and - treated stems wont null against the original But just to clarify, in case there is a misunderstanding, this was what I was actually discussing: 1) Null Test - original track (full mixdown) - SL10 unmixing the track, creating stems - taking all the SL10 extracted stems into the daw - taking the original track in a daw - playing all SL10 stems at once, together with the original track - flipping phase on original track, while playing - result: silence, null -> good thing, cause no artefacts created during unmix operation in the stems 2) Remix, Mastering, Restoration use case - unmixing a track in SL10 for having it separated into stems (- removing noise or unwanted stuff from a live recording in SL10) - rendering out the single stems with the edits - taking all the edited stems into daw - processing stems further on with fx, gain manipulation, etc but only slightly more in terms of mastering, depends whats possible - in my case reducing room reverb, bringing vocal more into focus, balancing the track - rendering everything out into a stereo track
Not only does SpectraLayers 10 unmix drums/bass/vocals/instruments it also splits drums into cymbals/kick/snare and then you can split vocals into main and backing so you could end up with 9 stems in stereo or 18 if you split left and right channels. Nothing and I mean NOTHING comes close and there is no competition. SO you unmix main track then next take the drums stem and unmix it and then the vocals and unmix that to. Whilst its not perfect it does a DAMN GOOD JOB !!!
Spectralayers 10 is the cleanest with the least artifacts. But despite that RipX sounds more like the original. It depends what you're planning to do with the stems. For AI voice training I would still prefer the cleanest version from Spectralayers 10. For remixing I would prefer the RipX version because you could work around most artifacts. Thank you for the comparison and have a nice day!
One more thing is that all the rebalancing of you mentioned except SpectraLayers will not null against the reference track if you put all layers together This is not important for one who use the algorithm to re-mix only But if you plan to use this for something else where you use all layers generated then I have more confidence to work with spectralayers Also good to mention RipX which is faaar better than acon and Rtx and in some layers will do better than spectralayers but no null here if put against reference track Also there are open source denoising / remixing AI models you should explore like demcus, onx, VA And believe despite being (open source) it perform better (far better) than some premium alternatives in some use cases
Thx for all the tips! RipX once contacted me, but the demo they gave me expired and we had rather a weird talk back then, so I quite avoid them ;) Cheers!
Ah, kind of same findings you have, could have spared me my writings, if I would have read yours earlier :) And yes, also context dependend what's best prog for a specific task, right
@@thisguy5611 not only links, but a mentioning of certain resources makes my message disappear. I've rewrited it like 2 or 3 times now, it's about a gcollab model, letting people use newest demixing models in their browser, without need of a good GPU. Maybe I should email mr. Paul instead..
Superancko, dzięki za komplement! Spectralayers potrafi o wiele więcej niż tylko rozbierać materiał na warstwy - dlatego tyle kosztuje. Może kiedyś zrecenzuję go jako całość, ale to byłoby naprawdę dłuuugie wideo...
SpectralLayers did the worst job at seperating layers. Listen to the start of the vocal include that white noise sound that the others tried to remove. RX10 at least gives an option to control the degree of separation. This lack of seperation degree in spectral layers actually gives it less artefact but less definition. Lalala was by far the worst
True, but my experience with their marketing people is very so-so, so I decided to skip it. They didn't give me an NFR, they simply sent me the trial and said I can make a test using it. Then the trial expired, because I had other vids to shoot. Happens!
He doesn't care, the goal of this video is clickbait. The best solution, which is free, is not even talked in the video. Meaning that this chan makes video about things he doesn't know anything at all about it, or he's just a liar and a clickbait whore telling you that dog shit is gold.
Heeey! How do you like Spectralayers 10 job? Planning to get it? Lemme know down below, thanks!
I use SL10Pro, now we got 10.0.10 with GPU support - check this :)
I also use Acoustica Premium and RX9. I prefer SL10 for unmixing, but not impressed about de-noising in this tool.
@@GadesChannel you should give de-noising another try, it's probably the best of all 3 in that area as well. Can't link here, but I've set up a page comparing SL10 vs RX vs Acoustica. google: divideconcept spectralayers restoration comparison
I´ve compared RX with Spectralayers. Spectralayers wins. Highly recommend - what a great tool!!
Same feelings here!
Thank you! Was looking forward to this one!
No problem, Sam. Thanks!
Exactly what I was looking for. thanks so much!
Happy to have helped you!
Welcome to Warsaw Paul Land.
Hey man. Hope you are doing well. Watching this now and have to say i regret buying teh Acon Remix plugin.
I does nothing my Rx cant do. Only its handy sometimes to have it in the daw as a plugin but in reality i never used it there.
I would have liked to hear just the rendered separated parts played back together, like all the Spectral Layers pieces played together to see if it adds back up to the original. Fantastic job comparing them!
Thanks, Jim!
SL10 extractions played together perfectly null against original track, see my comment above
I didn't even give Spectralayers 10 a chance when they announced it. Thanks for the review, it sound really good - unexpectedly good :)
Right? First version of a new module and already so good!
curious, why didn't you give it a chance initially ?
@@rldivide I tried v7, 8 and 9 and was rather unimpressed
@@glennloopez when it came out v7 had the best stem separator as well. But it wasn't improved until v10.
It's called Slayer for a reason
Great show Stay Cool My Friend
Thanks, James!
BTW Spectralayers 10 can then process your Drums stem again to split it down into the different percussion stems.
This is quite insane. Thanks!
Did you ever finish your template for Metagrid for cubase. I was really interested in buying it.
Yes, I did, I'm about to finish the documentation next week and send it to Metasystem.io for review :)
@@FoliaSoundhaha that is great news!! Looking forward to it!
Good Video 👍🏿👍🏿👍🏿
Thx!
Wow! Awesome! You made me want to get my hands on spectralayers 😁
Nice to hear that. Give it a go, totally!
Very good video, thanks much!
i own spectralayers. its pretty cool!
Hey, as a since version 5 Ableton Live user, how's Cubase Like these days?? I started on it. Can it do what Ableton does ?
They're different beasts. Liveacting or DJ sets - only Ableton. Deep audio editing - only Cubase :)
I have Cubase 11 Pro, and my spectralayers doesn't have an eraser tool. Anyone knows why?
Sorry to hear that :( I bet switching to Cubase 12, even the trial version, should be the first step to check things out!
When listening to the drums of the SPECTRALAYERS, you can hardly hear the shakers, although the RX has them. But in terms of sound quality, SPECTRALAYERS surpassed everyone in everything.
Which stem extractor software is the best for piano parts extraction from audio songs? Thanks!
Go with Spectralayers and see what happens!
@@FoliaSound Spectralayers is not the cheapest, but if the piano separation is the best than i will buy it.
Thanks.
Please have a look to RipX as well for stems separation in particular in layers
I've had their demo, but it expired. They contacted me once, but it wasn't my best experience.
@@FoliaSound Okay. Bad to hear. RipX primarily aims at DJ's and producers. Stems creation and sound(s) replacement and arrangement. Where I think RX and SL focuses on sound designers and sound restoration. BTW: I am not linked to RipX.
What about ripx?
RipX people contacted me once, but they didn't seem very cooperative when talking about reviewing RipX, so I never went for it...
yes Spectralayers, is by far the best and the cleanest one.
aaand... Dom Sigalas showed also that the extracted stems played together and tested nulling against the unmixed original track, that there is a perfect null with spectral layers!
Same I see with own tests and that is simply awesome.
I also tested this with HitnMix RipX but does not null.
Frankly - why would I need it? I need it to extract proper steams for further use. What's the advantage of the stems summing up to a null? What's the practical advantage of this? Is it good at all, because it means, that same frequencies are split and shared throughout different instruments? Doesn't it make them a bit hollow and empty, a bit like mp3 encoding? Serious question!
@@FoliaSound
Dom Sigalas reffered to Mastering to see an opporunity there using it.
Or kind of remixing, Dom put a delay for example on the voice, which wasn't there before and it worked perfectly.
Also reverb, in general, is used in mastering sometimes, so if you can go a little more surgical, why not?
Also when there is only a final mix left and the mix session or stemms aren't available anymore, nice to have a null.
I for myself had some live recordings (single microphone, voice plus one instrument) where there was some room noise and talking and stuff and which I had then easily on (a) layer(s) which i could simply edit and/or delete, quick delivery was needed.
So you can see it also in context of a restoration use case, if you will.
Afterwards I was able to render it out merged back again. And when the extracted stems are in such a quality that they null gainst the original, bam!
Same story with RipX, but I had artefacts being introduced in some cases.
And comparing RipX against SL10 (which you hadn't in your test, I read about the reason), RipX would not have kicked SL10 from the throne, but quite often would have been on par.
@@FoliaSound In the extraction process, yes, frequencies are always shared between insturments, thats where a quality algorithm is needed, that cuts out precicely, but only what is really needed, a really surgical clean cut, that keeps part alive and musical, thought about that too, but...
What I can say, at least, that I had that empty and hollow feeling a little with RipX, had the feeling like a little bit of liveliness was sucked out, got a little dry, dont have that impression with SL10 so far.
But both were often, in my tests, quite on par, just nuances.
I guess it also has to do with the quality of the track you want to split (mixing, format), obviouly.
But what RipX and Sl10 both weren't able to accomplish was extracting the bells out of the Smashing Pumpkins song "Disarm".
But anyways, SL10 makes best musical separation and leaves you with a null if necessary, whats not to like about it?
@@cpssounds3656 , I believe, that nulling when reverse separating is actually not a good thing and I can easily explain it to you. The fact, that separate instruments of your mix share frequencies over time doesn't mean that when you render them solo, they'll null when being added up, right? They create a mix because of all the routing, harmonic content and tonal and dynamic processing.
When dissecting the track, using SL10, it's different - you separate from the sum of sound that you have, and then, in reverse, you simply sum the channels up. No routing nor any dynamic/tonal/harmonic work and character is recreated. Thus rendering a solo instrument from you master will always sound better and fuller that something that nulls. When it nulls, it means it lacks a lot of info about the instrument and its real content that has been masked in the master, but it's still there.
In SL10 it's not masked - it's just not there. I'd realli like to see the elements of separation not nulling in SL, which would give me the hope, that this algorithm tried to retrieve something masked instead of sharing the master it had at the input. This is really not the best way to recreate the source instrument.
Hope it's 100% clear on my side now!
@@FoliaSound
Ah, I guess here is the culprit, that maybe brought in some confusion:
"Afterwards I was able to render it out merged back again. And when the extracted stems are in such a quality that they null gainst the original, bam!"
- "Afterwards", meaning after i removed room noise, talking and coughs and stuff, no fx, eqs or stuff
- "renderign out merged back again", meaning only the cleaned stems (without the noise) rendered back into a single stereo track
- "...they null against..." meaning, you have less artefacts in such processes, when the untreated stems alone null against the original, generally speaking
- of course the extracted - and - treated stems wont null against the original
But just to clarify, in case there is a misunderstanding, this was what I was actually discussing:
1) Null Test
- original track (full mixdown)
- SL10 unmixing the track, creating stems
- taking all the SL10 extracted stems into the daw
- taking the original track in a daw
- playing all SL10 stems at once, together with the original track
- flipping phase on original track, while playing
- result: silence, null
-> good thing, cause no artefacts created during unmix operation in the stems
2) Remix, Mastering, Restoration use case
- unmixing a track in SL10 for having it separated into stems
(- removing noise or unwanted stuff from a live recording in SL10)
- rendering out the single stems with the edits
- taking all the edited stems into daw
- processing stems further on with fx, gain manipulation, etc but only slightly more in terms of mastering, depends whats possible
- in my case reducing room reverb, bringing vocal more into focus, balancing the track
- rendering everything out into a stereo track
Not only does SpectraLayers 10 unmix drums/bass/vocals/instruments it also splits drums into cymbals/kick/snare and then you can split vocals into main and backing so you could end up with 9 stems in stereo or 18 if you split left and right channels. Nothing and I mean NOTHING comes close and there is no competition. SO you unmix main track then next take the drums stem and unmix it and then the vocals and unmix that to. Whilst its not perfect it does a DAMN GOOD JOB !!!
Same feelings here. Cheers!
Yes Spectralayers 10 sounds cleanest to me too.
But each of these are great.
I actually use Spectralayers 7 still only to analyse Songs.
Best Regards
Wow, this is cool. Greetings!
Spectralayers 10 is the cleanest with the least artifacts. But despite that RipX sounds more like the original. It depends what you're planning to do with the stems. For AI voice training I would still prefer the cleanest version from Spectralayers 10. For remixing I would prefer the RipX version because you could work around most artifacts. Thank you for the comparison and have a nice day!
Thanks! Yep, I've heard good stuff about RipX, but I got bad experience with their marketing people ;)
One more thing is that all the rebalancing of you mentioned except SpectraLayers will not null against the reference track if you put all layers together
This is not important for one who use the algorithm to re-mix only
But if you plan to use this for something else where you use all layers generated then I have more confidence to work with spectralayers
Also good to mention RipX which is faaar better than acon and Rtx and in some layers will do better than spectralayers but no null here if put against reference track
Also there are open source denoising / remixing AI models you should explore like demcus, onx, VA
And believe despite being (open source) it perform better (far better) than some premium alternatives in some use cases
Thx for all the tips! RipX once contacted me, but the demo they gave me expired and we had rather a weird talk back then, so I quite avoid them ;) Cheers!
Ah, kind of same findings you have, could have spared me my writings, if I would have read yours earlier :)
And yes, also context dependend what's best prog for a specific task, right
I want to second the call for you to test Ultimate Vocal Remover, it's super good
I'll try it then, thanks!
..its amazing, it would be nice to compare to FADR too...regards
Why did you delete my comment? about UVR5
I don't remember anything like this, sorry...
Write it again, it seriously wasn't me!
@@FoliaSound no, I'm sorry, guess that youtube really has bad relations with links
Depends on what you're trying to post.
@@thisguy5611 not only links, but a mentioning of certain resources makes my message disappear. I've rewrited it like 2 or 3 times now, it's about a gcollab model, letting people use newest demixing models in their browser, without need of a good GPU. Maybe I should email mr. Paul instead..
SUPER KANAŁ. Jak zwykle obiektywny, wyczerpujący i świetny materiał. SP10 Pro robi robotę ale 1330,39 zł czy to oby nie za dużo?
Superancko, dzięki za komplement! Spectralayers potrafi o wiele więcej niż tylko rozbierać materiał na warstwy - dlatego tyle kosztuje. Może kiedyś zrecenzuję go jako całość, ale to byłoby naprawdę dłuuugie wideo...
The song is so overcompressed!!!! Its is pumping so unnaturally that it makes my cans fart.
Spectralayers 10 vs RipX should be close, but RipX should win plus it can do a lot more besides.
SpectraLayers can also do a lot more :) But they are different tools anyway.
SpectralLayers did the worst job at seperating layers. Listen to the start of the vocal include that white noise sound that the others tried to remove. RX10 at least gives an option to control the degree of separation. This lack of seperation degree in spectral layers actually gives it less artefact but less definition. Lalala was by far the worst
I need to see a RipX comparison
Try UVR 5.5.1 (it's free).
I will, thanks!
It doesn't provide as many stems, is pretty destructive and doesn't have any tool to correct AI mistakes...
I get your explanation, but omitting RipX makes this video much less useful.
RipX might have been a more interesting contender than RX and Accustica.
True, but my experience with their marketing people is very so-so, so I decided to skip it. They didn't give me an NFR, they simply sent me the trial and said I can make a test using it. Then the trial expired, because I had other vids to shoot. Happens!
He doesn't care, the goal of this video is clickbait. The best solution, which is free, is not even talked in the video. Meaning that this chan makes video about things he doesn't know anything at all about it, or he's just a liar and a clickbait whore telling you that dog shit is gold.
ddd