The new AI is really wild. We are now getting some pretty convincing vocals. Freddie Mercury doing Billie Jean. AI simulating an early Paul (80s) doing 2018's I Don't Know from Egypt Station, which features a much older sounding Paul A couple thoughts... -Many of these are only sort of convincing now, but in 5 years...? -What if this was used as a tool. What if McCartney recorded a song like Fuh You, and de-aged his own vocals using AI? Legit?
If I had a million chances to guess the singer, Freddie Mercury wouldn't have been one of my guesses.
A pretty good Freddie, but in that first verse, I swear that he’s singing flat. Is that even possible?
Right now, it's too unlistenable to be convincing As a way to manipulate vocals on a modern song I say why not. Specifically if you have an amazing catalog of untouched vocals to work with, it could be incredibly useful for making new songs - it's not like studio trickery is new, so I can't say I would care too much if the technology was there
We were discussing this on another thread, but hearing the AI Paul singing "I Don't Know" is mind-blowing -- by far the best I've ever heard. If I was McCartney and I was aware of this, I'd use AI for all my vocals in the future. No kidding. It's that convincing.
I agree, it is shockingly good (at least on my IPad). Looking forward to hearing other forum members’ opinions! I’m not sure I can grasp all the implications of this tech yet, much less judge it as good, bad, or mixed: AI generated “new” songs, “sung” by people’s favorite artists, living or dead, sounding like they’re in their prime…even if the songs are not “written” by AI, it still feels revolutionary.
Somebody needs to convince Yoko to give Paul the rest of John's unfinished demos. Paul can finish them off so we have a bunch of new Lennon-McCartney songs. Let the technology improve a bit, and then give us a final Beatles album.
Does this mean that one day we'll get a version of the Yellow Submarine movie with convincing Beatles voices synced to the animation? The John and Paul voices in the original movie were close enough to pass, but the George and Ringo were anything but.
At this point any "new" discovery will have to be looked at with very careful ears and have provenance to back it up. Problem is. Do new generations even care?
I can assure you these A.I. possibilities have not escaped the notice of corporate powers-that-be. Pretty soon all the griping about Auto Tune will seem quaint, when Apple grants Giles Martin et al permission to create a new Beatles album. Queen making a new album with "Freddie". Or something similar. Think it can't happen? Off-topic but, my first big fear is what might happen in the 2024 U.S. election with deepfaked opposition ads. I know that's already being looked into by the advert providers for both parties (and the Chinese, Russians etc.).
This will cause a few heads to explode in this forum, but I’ve come across large amounts of “Kanye” covering Taylor Swift. Much of it sounds really good. or maybe it’s actually Kanye.
My 2 cents. Paul's take is really , really good. However, I want to mention that the term "AI" is starting to be used a bit too freely (and I am referring to the Youtube posters who put the titles). AI refers to activities which require a prediction, a classification, or any task which requires decision making (data driven). This application is completely "deterministic" (meaning, there are no decisions once the created model is set). This great example (Paul singing as 40 years ago) is more a case of very advanced digital signal processing to synthesize a vocal tract (the area from the nose and the nasal cavity down to the vocal cords deep in the throat). Obviously this is brought to reality given the advanced computing capabilities of today. Every singer's vocal tract can be modeled as a filter, that transforms each musical note's fundamental frequencies into a complex waveform with many harmonics (the same happens with musical instruments). What I imagine they are doing is that they are estimating the vocal tract filter by training a computer with many examples (probably most of McCartney's catalog), then "demixing" the vocal track from the song, putting that signal through the filter, and remixing it all again. This is not unlike the noise reduction techniques that have been used for decades whereas you transform a signal by estimating a "filter". Once the filter is calculated, there are no decisions and everything is just deterministic. That is why in my mind this is more a case of advanced DSP, as opposed to AI.
Prior to hearing that McCartney AI vocal, I would have been against something like this. But if we had a new Beatles album -- totally convincing vocals from John and Paul, playing that is indistinguishable from the honest-to-God Beatles -- would that be so terrible?
My thoughts? We should all be scared. That's my thoughts. Because I will guarantee that right now, some guy in a suit is thinking hmmm... hire song writers, and when this stuff matures, I can sell all new records of all the dead greats! And then, when the AI gets better at writing songs, I can employee only a computer and bang... money in the bank. AI will eventually destroy the arts if we let it, and right now, it seems we're all ready to let it.
I'm conflicted. The real test would be the writing. Tall, tall order for a machine. But I'm not betting agin it with my own money ...
Exactly. The writing of songs per se can be labeled as an "AI" activity as it pertains making decisions (chord sequences, lyrics, etc).
That's why I want Paul to work on John's unfinished demos now. Leave us something for posterity that we can work with, Macca!
That would be awesome : ). In fact what you described reminds me of when I first heard "Free as a Bird". I remember many people did not like it but I could not believe I would ever see the launch of a "new" Beatles song in my lifetime. That is why I am so fond of the Anthology series : )