A demo recording is a way for an artist to explore his ideas for a song, and/or to give to his backing musicians to learn a song’s riff or chord progression. On a demo, often he’s not singing with as much effort as he would give a polished recording designed for wide commercial release.
Unimpressed with the quality of these two ‘reunion’ tracks, Harrison declined to have anything to do with Lennon’s third demo, which consequently remains in its original state and was not included on Anthology 3, as originally planned.
As an application of artificial intelligence, making a badly recorded vocal performance “pure” is fairly benign — equivalent in some sense to cleaning dirty tapes, remixing a recording for stereo, or remastering it for new media. But AI does not — and it will not — end there. For both audio and video, AI can be used as a simulation tool, which an operator can use to sample a given voice or physical appearance and then create something entirely novel from the results. If we are not there already, we are getting close to the point at which a director could make a movie with a deceased actor in the lead role and at which a record producer could generate a full album’s worth of material “by” an artist who is now unable to record a note.
You may recall how in 1997 Fred Astaire's moving image was used in a series of Dirt Devil commercials, showing him dancing with the brand's vacuum cleaner. There was outcry against that at the time, and that's nothing compared to what's on the horizon.
For the Beatles project, Sir Paul says there's nothing fake about the AI used on Lennon's voice -- but he could say that honestly even if the AI used a pile of Lennon recordings, stealing a morpheme here and there and combining them to make Lennon sing the Malaysian National Anthem or whatever else you want. It's John's voice, just "remixed." Deepfakes are showing us how it's done, using existing images of famous people and merging them to make videos. A voice may be that much easier to fake with the actual sound of a person's voice, given a large enough dataset.
So, in effect, the upcoming song required unheard-of amounts of electronic fooling, was rejected the first time around by the now-deceased George, was something Lennon made three years before he was killed and apparently didn't think it met his own standards for publication, and will be mixed not as a Beatles song would have been done but as a McCartney (and I guess Starr) song.
I'm not terribly excited.
And look, if we're going to start screwing about with Beatles stuff using artificial intelligence, how about correcting some past problems? For example:
🎵 "Within You Without You" -- Did anyone not stoned ever enjoy this five-minute brick in the middle of Sgt. Pepper? I much preferred Big Daddy's take, doing it as a beatnik bongo number. It was still boring, but it was less than two minutes long.
🎵 "She's Leaving Home" -- Always wanted a more realistic take on the last verse, like:
Sheeeeee... is lying... in a diiiiiiiiiitch...
🎵 "Revolution 9" -- Could be replaced by something more interesting. Like, a blank soundtrack, or an apology.
🎵 "Sun King" -- Maybe we could speed it up a bunch? Might be a passable song played at 45.
🎵 "Boys" -- It has never been clear to me why Ringo sang this song by the Shirelles on the Please Please Me album. Maybe we could dub in the Shirelles singing. Or just dub in their whole song.
You laugh -- ha! ha! -- but it seems that AI is destined to be used to exploit the past -- and to "correct" errors of the past, the way censorship is being used to "correct" undesirable things in old movies and books nowadays. If our barbarous overloads are determined to destroy the past, maybe we could at least make it sound more pleasant first.
[refuses to take bait]
ReplyDeleteaww, come on
ReplyDelete(takes bait)
ReplyDelete(doesn't reply)
;>