Hey, a new sub-board!

[Lord Reith]

I like it. This is destined I think to become the "back room" where all the cutting edge stuff comes into being.

Back to work, ye boffins of AI electrickery!

[yymca6]

If this AI fad gets to be the real thing here, you can count me out. I'll stop re-upping stuff and will will delete all my "permanent" links.

We need the real thing, not these so-called wonders.

[alphabeatles]

Maybe we need a baseline rule: If an AI vocal replacement sounds anything like the Rutles or the Beatles cartoons, it doesn't make the cut. Don't post it.

So far I've heard exactly TWO AI vocal replacement tracks worthy of keeping: I Don't Know and My Valentine. Both feature the same singer in both original and replaced states (Paul), which may account for the fact that it actually works.

I do love de-mixes, not a problem, but that is "old" tech now

I don't think we can hide from AI, it's everywhere (photo editing, etc.). Just like people couldn't hide from any new "X" technology over the centuries. Not that everything is "good", there are some horrifying capabilities afloat at this time and usually used against us...

PS: Has anyone tried having AI write a new Beatles song (lyrics, vocals and music)-?

I decided to search and indeed some have, though I would never confuse this with a real Beatles song:

hxxps://www.flow-machines.com/history/events/ai ... pop-music/

?? hxxps://aitelegraph.io/2022/12/13/the-beatles-new-ai-album-across-time/

[Fast lucky]

I think the new sub-board is the one that was requested. I have nothing against AI demixes. They can bring new hindisght into songs. It is interesting and I never read any negative reaction against. Same for the remixes, if they are cleverly done, it can bring a song in a new light or present how a dropped idea would have sounded like.

The reaction from the members came about the fake - AI created vocals. This doesn't bring anything. Some might have fun doing it, fine for them, but it will only creates problems in the future about the check of accuracy. I listened to some 'projects' for the curiosity but I am not even as enthusiastic as aplhabeatles. Yes, I Don't Know sounds like Paul, but he would never have sung it that way with a young voice.

The only useful use that I can see is to repair a damaged acetate. There AI can trully help to recover a lost treasure.

All this to say that the requested sub-board was be about AI created vocals.

[urasam2]

If this AI fad gets to be the real thing here, you can count me out. I'll stop re-upping stuff and will will delete all my "permanent" links.

We need the real thing, not these so-called wonders.

I don't know what you mean by "the real thing". You don't seem to have a problem with fan made remixes - those are surely not "the real thing", so I don't see the harm in taking it a step further. Personally, I am loving the current spate of "AI Paul" songs. Sure, most are not very good but there are some genuine delights, and I am deriving a lot of pleasure in being able to hear them.

[stereoman33]

Having it all in a new area is the way to go.
My only issue is that it might overload the “Recent Topics” section so much that we might miss a new post of stuff like Melvin’s BBC stereo, a rare or new video , or a new live recording.
Id prefer it if it just stayed away from the recent topics section

[Juanco]

Hey all, big time lurker here.

I've been devouring these AI things for a week now. No offense, but almost all of them are sort of trash. So far I've saved a few of these creations, and one thing I can tell is that the ones made with just applying a skin of singer "x" over the original vocal track fail more than they are accurate, not only because of the occasional glitch, but because it sounds evidently false.

Now take this: Woman by John Lennon with AI McCa singing harmony hxxps://www.tiktok.com/@faulmccartney/video/7227203338780101931

For me it works. The creator sang actual new lines, imagining what Paul would have made, and then applied the AI Paul "skin". One could even make the case for artistry there (you need to both have the vocal skill and the knowledge on how the artist you're imitating sings, what he/she would do, etc.), although I won't enter that debate.

So yeah, I think that for interesting stuff, the technique where you sing imitating singer X and then applying AI on your own voice is the way to go –perhaps the only way to go, although sometimes the easier technique of doing it with the original vocal track works (I'm thinking George on Yesterday works, for some odd reason).

On another note, personally I don't get much of a kick of Paul singing Nirvana or some other random band. I think the community should focus on post-Beatles stuff and try to make some sort of a "new" illusory Beatles album. Also, changing singers (George on Yesterday, etc.) for me is interesting too. Oh, and one other thing: the Apple Jam tracks are cool, but I don't know, they sound obviously recorded in the 2000s, so the illusion is sort of broken from the start. Would be cool having a band/person recording like it's the 60s and then applying AI to their voices. If you go to @faulmccartney's tiktok, you can find some fake demos that play that trick.

PD: FYI, here's a channel that is aiming to remake Noel Gallagher's solo career with Liam's voice, doing it by that "singing your own lines" technique. Sounds cool so far. hxxps://www.youtube.com/@StarShaped_Tambourine/videos

[harrythebannister]

If this AI fad gets to be the real thing here, you can count me out. I'll stop re-upping stuff and will will delete all my "permanent" links.

We need the real thing, not these so-called wonders.

Couldn’t agree more. I cannot get into this crap at all. In fact the myriad of remixes shared here do nothing for me at all. The only stuff I have any interest in is the mono to stereo stuff as this adds some sort of new clarity to what we have.
I guess it’s time to move on.

[quigatolah]

For me, there just doesn't seem to be any life in these AI renderings.

When digital recordings made their first appearance, many complained that they didn't have the warmth of analogue recordings.
In the end, that turned out largely to be an emotional response to something that was merely different from what we were used to.
Who knows? Maybe I'll come around one day, but this stuff touches me less than 28kbs muzak.

I do get a kick out of mashups synching, for example, John & Paul (real) vocals into one track.
Even when they are bad, I still like hearing them once or twice. But to me, so far at least, this AI stuff is the music they would play in hell if there was such a place.

When you talk about (AI re-)contruction
don't you know that you can count me out.

[sunnylew]

I think of the attempts that swap singers, or rejuvenate Paul's voice more along the lines of outfakes than anything else - interesting and perhaps enlightening, but not revelatory.

The use of this technology as a production tool for audio repair is perhaps another matter.

Has anyone come across attempts to use this to capture the timbre of instruments? This technology may only be tuned towards voice capture, but if not ....

BBC recordings could be low hanging fruit for a test - the same performers, on the same equipment, recorded by the same people all around the same time. Some of the poorer BBC recordings could possibly be brought back to life with enough decent stems for training from the well recorded performances.

I have a training environment set up that I could use to have a go if anyone could steer me towards some good stems. I'd need about five minutes of, say well recorded rhythm guitar, and a stem of a poor quality one and I could give it a test if anyone knows of a separated BBC release.