by miobrien on 4/18/23, 2:20 PM with 201 comments
by forrestbrazeal on 4/18/23, 2:36 PM
For the record, here's my understanding of how this "AI-generated song" was made:
1. A human wrote the lyrics
2. A human recorded the beat
3. A human recorded themselves rapping
4. An "AI voice filter" made the recorded voice sound like Drake / the Weeknd.
IMO, the real tragedy here is not some threat to Drake (he’ll be fine) but that some dude who wrote an original song had to pretend it was written by AI to get it heard.
by alexb_ on 4/18/23, 3:08 PM
by pjc50 on 4/18/23, 2:46 PM
by rockemsockem on 4/18/23, 5:16 PM
They are going further than saying "take this song down" and are trying to go after training on any copyrighted data. It's a massive snatch and grab and a bit of a bait and switch too.
Straightforward impersonation that presents an AI voice as any specific human shouldn't be permitted, and US law is pretty clear on that I think. But an AI vocals system that works like stable diffusion where you ask for a voice with certain vocal characteristics that has been trained on copyrighted music? That's a VERY different thing and is much murkier. IMO it should be allowed, but what's clear is that the law doesn't know if it's allowed yet (in the US anyway).
by notefaker on 4/18/23, 3:58 PM
https://m.youtube.com/watch?v=y7r6PAkFRfU&feature=youtu.be
The implications of this kind of technology are wild. Imagine seeing a tribute show where the lead singer runs their voice through a voice changer that matches whatever artist they’re covering.
by squarefoot on 4/18/23, 2:43 PM
by Sol- on 4/18/23, 3:20 PM
Beyond this specific occurrence here, I think impersonating someone else's voice or appearance should generally be forbidden, except for say satirical use cases of course (which should come with some disclaimer).
by l33tman on 4/18/23, 4:18 PM
In this case, what is interesting is not that someone can (and did) produce a "fake Drake" song but that you could synthesize a singing voice of your choice and use it to produce songs without finding a new singer or training your own voice in the desired direction. I'm sure someone will make a prompt-based AI voice where you can just say "nasal annoying out of tune male voice" or something and control it that way and you can keep tweaking until you get what you want.
End result is of course even more (perhaps debatable) high quality content but with less star-appeal. We already have that on Spotify etc, there are loads of "troll" companies having hundreds of virtual fake artists releasing albums and tracks already. This will turbo-boost that..
However just like with AI images, it also enables someone with less talent of drawing or singing to do creative works. That has to be pretty valuable as well.
by poisonarena on 4/18/23, 3:37 PM
Kan and Aidode(israeli AI music generation startup) made a duet with two dead singers:
Ofra Haza + Zohar Argov https://youtu.be/7ND1Pw6QD_0
got 200k play in a few days, not bad for hebrew only. Very convincing.
Completely lost in non hebrew speaking world (99% of the world), and just as impressive I think.
by jetrink on 4/18/23, 3:12 PM
I bring this up, because it seems to me that if we live in a legal environment where a song's "vibe" is now de facto copyrightable, there is no way anyone will get away with cloning someone's voice (outside of parody.)
1. https://switchedonpop.com/episodes/invasion-of-the-vibe-snat...
by extr on 4/18/23, 3:27 PM
by nonbirithm on 4/18/23, 4:26 PM
Ironically it's been shown that the fans can also be on the side of AI. Some of them will happily download or train a LoRA based on their favorite artist and use it with Stable Diffusion, often to generate images the original artist would never draw (because of limited commission slots/doesn't take commissions at all, lack of experience/interest or explicitly stating they will not take commissions for certain genres).
by 999900000999 on 4/18/23, 2:43 PM
I can imagine some independent artists being more open to allowing AI imitation.
by sethkim on 4/18/23, 4:32 PM
The only conclusion we're able to come to is some kind of procedurally-generated music that is more specifically catered to each listener. I'm not surprised this wave is starting to break and it's a great time for artists and legal teams to figure out how to work with this technology if there's any possible way to do so.
by say_it_as_it_is on 4/18/23, 5:45 PM
by machdiamonds on 4/18/23, 6:08 PM
by CPLX on 4/18/23, 2:35 PM
But the song sucked, as did the supposed AI contribution. I couldn't quite figure out why everyone used this as the example of how it's all over and AI will take over the world of music.
by cush on 4/18/23, 3:09 PM
Reverse engineering someone's talent and identity and using it to create new work under their brand and identity should be a crime
Hell, take it one step further. Non-consentual deepfakes of any kind should be a crime.
by sizzle on 4/18/23, 6:47 PM
by jrm4 on 4/18/23, 3:39 PM
The main reason(s) that this was not impressive were
- Drake is VERY easy to imitate. He's little more than a Lil Wayne impression himself.
- "The Weeknd" here definitely sounded like an imitation of the Weeknd, I'm fairly certain that if I wasn't told in advance, I would be like "something is way off, bad day at the studio? Not actually him?"
by stall84 on 4/18/23, 2:53 PM
by azraeldv on 4/18/23, 2:40 PM