I assume (maybe incorrectly) that most pitch correction was/is intended to not be noticed, and isn't in fact noticed by me, though I remember on ILM c. 2003 some people were claiming they could always tell. Of course, in contrast to "most pitch correction" was the "Believe" effect where people decided they *wanted* to sound treated, and *that* became a whole new game – am thinking (again without knowing) that Autotune superseded vocoders, since the latter (afaik) limited you to sounding futuristic, whereas Autotune could make you be weird and cracked or turn your voice into a prism or kaleidoscope.
I'm guessing that A.I. can create weird effects as well as Autotune can – perhaps you can program it to sound like Autotune! – but am skeptical that it can create *A.I.*-specific-seeming weirdness rather than just sounding "oh, the voice sounds treated," like we've known about ever since 1969 when John Lennon decided he didn't like the sound of his voice so dropped it into a vat of chemicals on "Come Together." Or maybe ever since the Chipmunks.
So I assume most A.I. music won't be noticed as such by me, and won't sound robotic or futuristic, or particularly computer-generated – and when it does sound computeresque, that'll be because it's imitating techno, etc., not because its A.I.-ness is particularly detectible.
I think the short term impact on AI for vocals is probably the creation of fake vocals using real voices that are indistinguishable from genuine recordings. But interestingly the only ones that don’t scan as tech novelties so far are the ones where the vocals already sound heavily processed (like the GrimesAI). And it’s not that these faked AI vocals sound bad per se (though they’re still a little wonky), it’s that no one has figured out what to do with them and I don’t really understand the value (or the problem) with AI knock off voices except the usual concerns around deepfakes and malign actors. (“That’s not really Drake singing -- so what?”)
Am guessing that if I were still trying to create music there'd be a day when I could ask the bot to synthesize T-ara circa 2012 and MC Pipokinha – I don't just mean adding Pipokinha's vocals to a T-ara-sounding tune, but rather ramming together, say, her producer MT7 sound with their Shinsadong Tiger sound. Anyway, ask it to produce a set of songs, and I wouldn't necessarily get good music but maybe'd get good ideas I might not have thought of myself.
I could get the same from humans, of course, except they'd probably refuse, unless I had a lot of money.
And none of it will come close to what RIGHT NOW producers like DJ Will DF and DJ Jeffdepl (and many others) are doing just by snatching already extant vocals and running them through their noise machines.
I assume (maybe incorrectly) that most pitch correction was/is intended to not be noticed, and isn't in fact noticed by me, though I remember on ILM c. 2003 some people were claiming they could always tell. Of course, in contrast to "most pitch correction" was the "Believe" effect where people decided they *wanted* to sound treated, and *that* became a whole new game – am thinking (again without knowing) that Autotune superseded vocoders, since the latter (afaik) limited you to sounding futuristic, whereas Autotune could make you be weird and cracked or turn your voice into a prism or kaleidoscope.
I'm guessing that A.I. can create weird effects as well as Autotune can – perhaps you can program it to sound like Autotune! – but am skeptical that it can create *A.I.*-specific-seeming weirdness rather than just sounding "oh, the voice sounds treated," like we've known about ever since 1969 when John Lennon decided he didn't like the sound of his voice so dropped it into a vat of chemicals on "Come Together." Or maybe ever since the Chipmunks.
So I assume most A.I. music won't be noticed as such by me, and won't sound robotic or futuristic, or particularly computer-generated – and when it does sound computeresque, that'll be because it's imitating techno, etc., not because its A.I.-ness is particularly detectible.
I think the short term impact on AI for vocals is probably the creation of fake vocals using real voices that are indistinguishable from genuine recordings. But interestingly the only ones that don’t scan as tech novelties so far are the ones where the vocals already sound heavily processed (like the GrimesAI). And it’s not that these faked AI vocals sound bad per se (though they’re still a little wonky), it’s that no one has figured out what to do with them and I don’t really understand the value (or the problem) with AI knock off voices except the usual concerns around deepfakes and malign actors. (“That’s not really Drake singing -- so what?”)
Am guessing that if I were still trying to create music there'd be a day when I could ask the bot to synthesize T-ara circa 2012 and MC Pipokinha – I don't just mean adding Pipokinha's vocals to a T-ara-sounding tune, but rather ramming together, say, her producer MT7 sound with their Shinsadong Tiger sound. Anyway, ask it to produce a set of songs, and I wouldn't necessarily get good music but maybe'd get good ideas I might not have thought of myself.
I could get the same from humans, of course, except they'd probably refuse, unless I had a lot of money.
And none of it will come close to what RIGHT NOW producers like DJ Will DF and DJ Jeffdepl (and many others) are doing just by snatching already extant vocals and running them through their noise machines.
Maud the Moth sounds nice but doesn't live up to her billing.