Good comparison of Suno online platform’s AI generation with new stem separation options, vs SLP v11 (Unmix song) from Lanewood Studios…
I tend to agree with all his findings/assessment. Check it out.
Will be interesting what surprises/improvements SLP v12 release has in store… 
SpectraLayers wins hands down in not granting “a worldwide, non‑exclusive, fully paid‑up, sublicensable … license to use, reproduce, store, modify, distribute … in connection with … improvement of our products and services, including … machine learning models” for every single bit of any audio you ever use their service to analyze, though.
I guess some people don’t mind granting sweeping and total license, including sub-licensing, of one’s own work.
goodness! well, much of mine is online free to do what anyone will…not that anyone cares, only one guy involved with my musics on the planet actually worries about that stuff…when he remembers to fret about it
yeah, does this affect what Robin next? not that I viewed the video…yet
Oops.!
Well, yes - and I should have qualified, I’d never use such a service myself, let alone on any clients material (without telling them). Which is why I’ve invested in SLP.
No, I was just drawn to the quality of the stem separation comparison only and dreaming of what may be in store for us, with the upcoming SLP v12 release.
Thought others here may be interested, teasing the current ‘state of play’ with this algorithm technology…
Indeed, and it’s valuable! It’s just crazy what these companies are sneaking into the ToS/EULA terms isn’t it? Scary stuff for sure…
Yes, we’re sleepwalking into a time where ‘creative innovation’ will last only as long as it takes to scrape it into a LLM, for everyone else to exploit - and there’ll be absolutely no comeback around ‘stealing of ideas/using without permission’ etc, etc…
I bet Suno could now turn around and claim copyright infringement on the song that dude in the video used to test their own service and claim all income from that video.
EDIT: I didn’t watch the video yet, but I presume it’s his own content 
Could you explain to me more what he means by SUNO generating the stereo file and then separating the stems (ca. 9:45), i.e. “generate the music themselves” (ca. 15:00)? To confirm, does this mean in a technical sense SUNO is generating data in the stems that is not actually there based on an analysis of the stereo? I’m curious what this signifies forensically when stem extraction is performed to compare audio.
After watching the video, my impressions are mainly the following:
So any comparison to SL Pro seems extremely limited to this very specific narrow use case of Suno generated full mixes, and not with the use case of doing stem separation of arbitrary original material.
And I suspect, that Suno’s separation not passing the null-test, while SL Pro’s separation does also hints at Suno doing something different - potentially not actually separating the full mix, but rather something like “re-generating” individual stems.
If that’s the case, it would seem more comparable to exporting submixes from Cubase, which most certainly should beat the quality of even the best stem separation.
However, if you would upload your own original lyrics, then copyright concerns might indeed arise. So I definitely would not want to upload original lyrics I wish to protect from being ingested.
Side note: Giving someone the right to use my copyrighted material does not necessarily imply, that I lose copyright. So the conversation around Suno’s T&C may deserve more nuance.
p.s. Again: I’m not a fan of AI generated music, or even AI in general. But I do believe that criticism does better, if it is more precise/accurate/nuanced.
EDIT: To clarify some more: While I’m not a “fan of AI” in the gushing fanboi sense, I do consider it a technology that is positive for some applications and negative for others. Pretty much like any tool/technology.
Indeed, it says “non-exclusive.” But they can basically do whatever they want with it as long as they claim it “improves their service.” That doesn’t preclude you from claiming copyright, but you can’t against them, including if they use your music in their generative algos.
As I understand it, what the reviewer did was use the Suno AI Platform to enter some text description (stylistic prompts/desires) and some lyrics and then presses the button to get Suno to generate a song, given that set of input. He regenerated the results around 9 more times, each producing a stereo file variation using the same prompts. One of which he found appealing (to his taste/judgement/sensibilities).
He then chose to download this stereo file AND used the platforms latest offering of full stem separation files too (performed on that same stereo file, I’m presuming…).
Then he did his analysis on this stereo file and the zip download of the stems produced, in Cubase. He included a comparison with SLP’s attempts at stem separation of the same stereo file.
Hope that helps.
Correct on both counts.
You say ‘extremely limited’… Well, for me it’s still a perfectly valid ‘Unmix Song’ task for SLP to undertake - can’t be selective in choosing what’s thrown at it, surely.? Yes, SLP is a lot, a LOT more than that - but within the confines of the comparison experiment, the reviewer stated in his opinion SLP is one of the best at what it does (for other products to stand up against).
Yes, there may be something in that…
I’m not sure how to get from 
to 
It would also be a perfectly valid “Unmix Song” task to make SLP unmix a song of mine where I had the original stems myself.
Which is what I tried to express:
So why would I even run a comparison between an unmixing tool and having access to raw stems?
So I think the title of this thread is misleading, since we don’t even know if Suno is actually unmixing. I think the new title reflects the video test much better - thanks for editing it!
Because it’s on YouTube, and Suno now has the same rights to distribution. They could get at least 1/2 the monetization, but would probably just get all of it because the content creator would have to fight it at the risk of a strike, and possible channel removal.
EDIT: This particular example is different because the original was apparently Suno-generated in the first place, but any other original works example would apply.
On the other hand, there’s this:
That’s just saying if you use the Pro/Premier generation service you can use the “commercial use license.” That doesn’t preclude the granting of the right stipulated in the ToS when using other qualifying services like the online un-mixer.
Two different things. One should also note that copyright entitlements don’t necessary matter on YouTube insofar as claims / strikes are concerned. What that article discusses is kind of apples and oranges…
The important bit, and the reason I posted in the first place is this:
“By using the Service or otherwise transmitting Submissions to us, you grant to Suno and our affiliates, successors, assigns, and designees a worldwide, non-exclusive, fully paid-up, sublicensable (directly and indirectly through multiple tiers), assignable, royalty-free, perpetual, irrevocable right and license to use, reproduce, store, modify, distribute, create derivative works based on, perform, display, communicate, transmit and otherwise make available any and all Content (in whole or in part) in any media now known or hereafter developed, in connection with the provision, use, monetization, promotion, marketing, and improvement of our products and services, including the Service and the artificial intelligence and machine learning models related to the Service.”
If you’re going to use the service and plan to copyright content, then I suggest you carefully read this, and of course hire a lawyer
I wholeheartedly agree with that.
And yes, the TOS language is so insanely overreaching, that nobody should touch that service with a 10 foot pole, if they have any traditional copyright aspirations.
It’s despicable behavior, and it’s specifically designed to take advantage of people knowing they won’t read the ToS first. And of course they can pay a team to provide what is probably decent audio software services for free when people are literally handing over their intellectual property to the company every single time they use it.
100% agree with you, sir.
Apologies @Nico5 - I hadn’t fully appreciated your point here. Yes, understood; I see what you mean.
In weighing up any comparison of tools/processes such as this, we must ask if Suno is performing an actual ‘Unmix song’ separation on the stereo file.? Or, whether it’s just spitting out whatever stems were generated in the first place, in order to produce the final result.
And yes, 100% fact, raw stems from a DAW on one’s own material will be infinitely better than any stem separation process could manage.
EDIT - also now changed topic title.
We have been using AI text to voice for a couple years now (with our own recorded and edited talent and copy; not service presets) and the results have been stunning, to be fair…it has saved me many, many hours of mixing that I has been doing previously…and recording VOs is a thing of the past for many deliverables
Further, some AI generated art work fed from our photos has turned in some stellar results.
AI music?; no thanks