AI-Oasis are musics' Deepfake-moment
Supersonic.
Oasis have a new album out and it’s bonkers. Except it’s not Oasis, but a sonic deepfake.
The album was created by a real band named Breezer who changed the voice of their singer with AI to sound like Liam Gallagher, and it sounds pretty much like the great lost album that never came out after Morning Glory. Breezer are on Spotify, they already sound pretty much like Oasis in a good way.
From the Guardian story on this album:
“We just got bored waiting for Oasis to re-form,” says Bobby Geraghty, a 32-year-old singer, songwriter and producer. “All we have now is Liam and his brother trying to outdo each other. But that isn’t Oasis. So we got an AI-modelled Liam to step in on some tunes that were originally written for a short-lived but much-loved band called Breezer.” […]
Bobby Geraghty, explaining how he cut up various a cappella recordings of Liam to train his AI one [said:] “Obviously, our band sounded exactly like Oasis. So then all I had to do was replace my vocals with Liam’s.”
Two days ago i wrote about how that viral AI-song by synthetic Drakeweeknd sounds just as lame as the original, how it just is the same cheap stupid stuff these two produce, but even cheaper and more stupid, which is astounding because nothing is more stupid than Drakes music, except maybe that of Kanye West.
AISIS is the exact opposite: A real band deepfaking the voice of their singer to sound like, arguably, the best band on the planet from 1993-1996. From their first single Supersonic to their second album What’s The Story Morning Glory, no other band in the world created an comparable wall of sound.
Just listen to the Oasis-Livegigs from Glastonbury 1994 — then a well known band but not the sensation they are about to become —, and Knebworth 1996 and imagine what those guys did to Rock-music in those two years. Oasis went from unknown to Champagne Supernova in the sky in pretty much no time and these streetwise brothers and their artful arrogance was the best thing that happened to Rock and Roll since Nirvana and Grunge. Now, these Lost Tapes are definitely maybe the best Oasis-Album since those days of Morning Glory.
In true Liam Gallagher-fashion, he praises this album, mostly because it sounds like him:
Just as Liam, I'm impressed with the singing on this AI-album, not just because it sounds like him, or Oasis like you haven’t heard them since 1997, but also because it’s a technical achievement.
Drake and Weeknd are both are rappers, and many, if not most, AI-music going around right now is Rap and Hiphop. Which, fine, it’s not my sound and mainstream acts like those two are hardly the best example to choose — any Deepfake-Knxledge out there? —, but i'd also argue that rythmic speaking is easier to do with AI-voice-clones than singing.
However, I don’t think this is AI-musics Dall-E-moment, because this is not really generative AI spitting out synthesized stuff as a whole, like Midjourney spitting out a fake pope and everybody goes nuts. AI-voice-cloning already was mostly mentioned in Deepfake-contexts, and this is AI-musics’ Deepfake-moment.
As a pretty big fan of Oasis back in the day, i’m more than happy that the universe chose one of my alltime favorite bands for this moment in AI-history.
at compares this album and the state of AI-music to the release of the genre-defining The Gray Album by DJ Danger Mouse, which fused The Beatles White Album and Jay-Zs Black Album and earned a legendary status in the mashup-genre not least due to EMIs fruitless attempts at surpressing it.I ran a MP3-blog called Mashr and worked at the (already sued and legalized-as-a-streaming-service) Napster back then, and today this once illegal album is part of pop canon with DJ Danger Mouse being one of the most successfuls producers out there, who has been nominated for 22 Grammy Awards and won six. The situation around copyright and mashups back then reminds me a lot about the weird copyright issues surrounding that AI-Drake-song right now.
I like that comparison and In AI cinema gonna be wild, i already wrote about how AI is bringing back postmodern mashup-techniques on a much more sophisticated level while lowering technical barriers of entry:
the beatniks in the 60s used cutup techniques to create unique artifacts that kickstarted the postmodern revolution in art, but these techniques were exclusive to avant garde artists on the edge of contemporary thinking, or too expensive, to be used by the collective on a societal level. Digital Tools and price drops in hardware lowered that barrier to near zero, all you needed is technical skills. Then Social Media enabled us to select the best of those creative works for virality, spreading a contagious memetic creativity-virus that led to the mashup-craze of the 2000s.
Enter AI.
Suddenly, even that technical barrier is gone and anyone can create dozens and dozens of images from Jim Hendersons Labyrinth with H.R. Giger as art director.
For now, true generative music-AI is not there yet. Just like with visuals, where Deepfakes entered the conversation five years ago when computer vision and image synthesis was not ready for prime time, the current state of pure music synthesis is mediocre at best, lacking consistency and sound quality. Doing good stuff like The Lost Tapes of the 27 Club requires a lot of editing and consists of AI generating MIDI-files, no direct audio. Even with generative AI-tools which output direct audiofiles like OpenAIs Jukebox, quality and consistency still are subpar and the vocals remind me of the typography in image synthesis one year ago. Here’s AI-Nirvana as an example.
But i have no reason to believe this will not improve. While i think that music is harder to synthesize than visuals, because it has many more dimensions than images (time, rythm, consistency, song-structure, and more), i also think that music will soon have a Dall-E for tunes. The labels are already up in arms, and i keep repeating myself: I have no idea how you could, ever, regulate a technology that analyzes the structure of cultural expression, smashing it into atomic pieces that you could never describe as a Sample, and turning all of the resulting grey goo into an interpolatable latent space, explorable by three trillion billion dimensions.
What a time to be alive.
Now all I want from life to be happy is a podcast —like the infinite conversation between Werner Herzog and Slavoj Žižek—, in which Liam and Noel dunk on each other in Manchester slang eternally in a never ending feud between the coolest brothers in Rock and Roll history, you fookin’ twat.