Beatles song is created thanks to AI

[WillyW]

Summary

With technology from Peter Jackson (Lord of the Rings producer), the remaining Beatles members were able to upscale a previously recorded set of tracks for a song and produce the song with direct input from the two deceased members of the band. The sources were a badly recorded cassette tape 40+ years ago from John Lennon, and a 90s guitar recording from Harrison, both preserved and enhanced and combined as tracks in the new song.

 

Quotes

Quote

...Director Peter Jackson was working on his comprehensive Get Back documentary for Disney Plus. His team developed a technology that allowed them to take practically any piece of music (even ancient demos) and “split all the different components into separate tracks based on machine learning.” McCartney and Starr realized this was their opportunity to go back and give “Now and Then” the ending it deserved. “Now we could mix it and make a proper record of it,” McCartney said.

 

My thoughts

This would seem to be one of the cases where AI has benefits that most people would not have thought to think of. There are many upsides and downsides to this. Say you have a media library with very poor MP3 encoded bitrate files, sometime in the future you could use a tech like this to upscale the set. Now all you would need is some AI to write a CMS for you if wanted to be able to host this on a home NAS or offline storage. Downsides could be that in future estate holders could try to claim extended copyright on works and try to get the 50 year copyright term extended in law reducing the ability for things to go open source and a be used to create more unique works, and enrich a few people who do not contribute much to society.

 

Sources

https://www.theverge.com/2023/11/2/23943290/now-and-then-the-beatles-new-song-ai - Article

 

Edited November 8, 2023 by WillyW
Added the song and a video documentary

[TheLANguy]

21 minutes ago, WillyW said:

My thoughts

This would seem to be one of the cases where AI has benefits that most people would not have thought to think of. There are many upsides and downsides to this. Say you have a media library with very poor MP3 encoded bitrate files, sometime in the future you could use a tech like this to upscale the set. Now all you would need is some AI to write a CMS for you if wanted to be able to host this on a home NAS or offline storage. Downsides could be that in future estate holders could try to claim extended copyright on works and try to get the 50 year copyright term extended in law reducing the ability for things to go open source and a be used to create more unique works, and enrich a few people who do not contribute much to society.

I personally think media outlets saying the song was "created with AI" is misleading. Machine learning was only used to split the piano from Johns vocals and upscale them. That was it! Not very surprising when tools like lalal exist.

As a Beatles fan, it is great to see the song completed.

[Dedayog]

32 minutes ago, TheLANguy said:

I personally think media outlets saying the song was "created with AI" is misleading. Machine learning was only used to split the piano from Johns vocals and upscale them. That was it! Not very surprising when tools like lalal exist.

As a Beatles fan, it is great to see the song completed.

This.

 

Also, the voiceover industry is throwing a git about a tangential use of this tech.  May not need real  people to voice cartoons anymore.

[SansVarnic]

-= Topic Moved =-

This is more of a General Discussion type of subject.

[Lightwreather]

1 hour ago, Dedayog said:

May not need real  people to voice cartoons anymore.

I would actually disagree there. I mean, actually I genuinely hope it gets rid of all those terrible low-budget VAs that some studios hire, but Voice-acting imo is more for an art form, and thus the VA actually adds a dimension to characters that I doubt an AI could really replicate? To note: I am not particularly well-informed of the state of AI in voice-acting, I could be very well wrong. My experience only goes so far as a vocaloid and similar

[Needfuldoer]

2 hours ago, Lightwreather JfromN said:

I would actually disagree there. I mean, actually I genuinely hope it gets rid of all those terrible low-budget VAs that some studios hire, but Voice-acting imo is more for an art form, and thus the VA actually adds a dimension to characters that I doubt an AI could really replicate? To note: I am not particularly well-informed of the state of AI in voice-acting, I could be very well wrong. My experience only goes so far as a vocaloid and similar

AI text to speech isn't the biggest looming problem for voice actors, because computers still can't conjure convincing tone and inflection out of thin air. This is a pretty typical example. The voices sound like Thom Huge as Jon Arbuckle and Lorenzo Music as Garfield, but they're flat and robotic:

 

 

AI voice mapping is a whole other animal, because you can put together a fake vocal track with just some recordings of your target and a human who can do a reasonable impression of them. That's what all the "AI covers" online actually are. As long as the impressionist can sound 'close enough', the AI model can work its magic, and human expression makes it through to the final, passable product:

 

 

But so far, there's still only so much AI can do. If the voices are too different, the results are... 'off'. Look at all the "Sinatra AI covers" where his voice warbles, or stuff like this:

 

 

Liam Gallagher sounds like John Lennon as it is. Run him through an AI model of John Lennon's voice, and it works:

 

 

I'll bet they ran John Lennon's old demo tape through a John Lennon AI model. That would effectively clean it up, but of course still sound like him.

 

There used to be a great example of this, where someone took Paul McCartney's 2013 song "New" and ran it through an AI model of his own voice from his Beatles/Wings heyday. The AI convincingly "de-aged" his voice. Unfortunately the only copy I can find now is somebody else's re-upload where they added a ton of Beatlemania screaming so it doesn't sound all that different from the studio recording.  

Edited November 9, 2023 by Needfuldoer

[TheLANguy]

6 minutes ago, Needfuldoer said:

I'll bet they ran John Lennon's old demo tape through a John Lennon AI model. That would effectively clean it up, but of course still sound like him.

I very much doubt that.

Having heard the demo cassette and watching them clean up the tape in the mini documentary. (7:04) You can still hear a bit of the grain from the cassette.

 

[Needfuldoer]

3 minutes ago, TheLANguy said:

I very much doubt that.

Having heard the demo cassette and watching them clean up the tape in the mini documentary. (7:04) You can still hear a bit of the grain from the cassette.

I admit I listened to the single but only skimmed bits of the mini-documentary, so that was just conjecture on my part. 

 

Sounds like they were able to completely isolate his vocals from the piano track, which was always the best way to go.

[TheLANguy]

2 minutes ago, Needfuldoer said:

I admit I listened to the single but only skimmed bits of the mini-documentary, so that was just conjecture on my part. 

 

Sounds like they were able to completely isolate his vocals from the piano track, which was always the best way to go.

Agreed.

Sorry if I sounded rude, just a big Beatles know it all.

[LMGcommunity]

I wish it had more of a classic Beatles feeling with some hard stereo panning and having John's piano playing be higher in the mix later in the song, but it's extremely cool that this exists.

[TheLANguy]

3 minutes ago, LMGcommunity said:

I wish it had more of a classic Beatles feeling with some hard stereo panning and having John's piano playing be higher in the mix later in the song, but it's extremely cool that this exists.

I wish they would have kept the "I don't want to loose you" section from the demo.

It's at about 3:21