I never said that

November 5, 2016 3:52 PM

(Adobe) VoCo allows you to change words in a voiceover simply by typing new words [slyt]
posted by lmfsilva (31 comments total)

Simpsons did it.
posted by tresbizzare at 4:14 PM on November 5, 2016 [2 favorites]

Note, no closed captioning
posted by AFABulous at 4:43 PM on November 5, 2016 [2 favorites]

I'm already working on the script for the next season of Black Mirror.
posted by adept256 at 4:59 PM on November 5, 2016 [4 favorites]

Well this is going to bring Bad Lip Reading to a whole new level.
posted by Anticipation Of A New Lover's Arrival, The at 5:03 PM on November 5, 2016 [2 favorites]

However I also feel like I can't share this with my friends because the example he used made me uncomfortable. Did the hosts agree beforehand that it would be OK if he used his new toy to make it sound as though they were having an affair? Did the woman, in particular, agree to have her name inserted in place of the male host's wife? It just seemed like a really hinky choice of subject, when there are so many possibilities that wouldn't have involved issues of consent or female objectification. Cool tech, but a bit of a side eye for the presenter.
posted by Anticipation Of A New Lover's Arrival, The at 5:15 PM on November 5, 2016 [5 favorites]

Videogame characters will be able to call you by your real name now
posted by LogicalDash at 5:50 PM on November 5, 2016 [8 favorites]

I really love my wife Trump. This will become mandatory.
posted by adept256 at 5:57 PM on November 5, 2016

Anticipation Of A New Lover's Arrival, The: "Did the hosts agree beforehand that it would be OK if he used his new toy to make it sound as though they were having an affair? Did the woman, in particular, agree to have her name inserted in place of the male host's wife?"

The Adobe rep was editing a clip of Keegan-Michael Key, a different actor, into saying he kissed Jordan (Peele), the other dude on the stage. You might know them better as Key & Peele, of sketch comedy fame. The woman on stage wasn't involved in the bit.

Also, combine this with the equally awesome (and creepy) real-time video editing project Face2Face (previously), and you can basically produce convincing footage of public figures doing and saying anything.
posted by Rhaomi at 6:02 PM on November 5, 2016 [15 favorites]

you can basically produce convincing footage of public figures doing and saying anything.
This will not wendell.
posted by oneswellfoop at 6:29 PM on November 5, 2016 [5 favorites]

Ah, apparently I had no idea what I was talking about! Thanks for setting me straight.
posted by Anticipation Of A New Lover's Arrival, The at 6:35 PM on November 5, 2016 [1 favorite]

I've been improving voiceovers for years now. Just press MUTE.
posted by Sys Rq at 6:45 PM on November 5, 2016 [1 favorite]

Gets to work on Frank Rizzo reads The Lord of the Rings audio books, lawsuit be damned!
posted by juiceCake at 7:44 PM on November 5, 2016 [1 favorite]

This is... a little obviously-fake in the example, but I'm assuming there must be some intonation and inflection controls they're not playing with. It would t take much tweaking of that to make it sound seamless.

This is amazing. And scary.
posted by egypturnash at 9:34 PM on November 5, 2016

Also

LogicalDash > Videogame characters will be able to call you by your real name now

Well, by a name. "Futtbucker! Get over here and punch this load-bearing boss!"
posted by egypturnash at 9:38 PM on November 5, 2016

Also this is gonna open up some major cans of worms for voice actors. Train it on someone, start having them voice your cartoon for free. I wonder how soon it'll be until people start selling packages of training data, similar to how you can buy meticulously complete sets of grand piano samples?

Also: I wonder what happens if you feed non-vocal samples into it as training data. Can it create weird alien voices out of field recordings of busy streets, bird-filled forests, or raging thunderstorms?

And I wonder how long it'll be before you can speak into a mic, and have someone else's voice come out based on what you're saying...
posted by egypturnash at 9:43 PM on November 5, 2016 [3 favorites]

Also this is gonna open up some major cans of worms for voice actors. Train it on someone, start having them voice your cartoon for free.

Season 90 of The Simpsons is going to suck.
posted by adept256 at 9:51 PM on November 5, 2016 [2 favorites]

I, for one, welcome our new post-fact society
posted by Dr. Twist at 10:14 PM on November 5, 2016

It glosses the technology. The demo plays up cutting/substituting one sample for another without bothering to explain how a non-sampled utterance is sourced. There is no regard for the feat of sampling a syllabary to produce everything, such as what was attempted for Marlon Brando and shown in Listen to Me Marlon (2015). Which is wonderfully spooky.
posted by lazycomputerkids at 10:44 PM on November 5, 2016 [1 favorite]

Also this is gonna open up some major cans of worms for voice actors.
Not until the software is able to do tone convincingly. Sure, the software might be able to splice phonetic syllables perfectly, but to actually threaten voice actors, it will need to do more than regular speaking voice.

I can see it becoming big in videogames. No more list of names on sports games, mods can have narration and characters with generated speech, for a start. You can have your RPG party of Butts, Farts, Skunkbreath and Peniswine call each other that, too.
But for that to happen, they are still going to need someone to give that program a library to work from. This tech seems great for patchwork, and correct a thing here and there, but it's likely far away from being a total replacement. But I do wonder if higher profile voice actors will start future-proofing their contracts to prevent being ripped off when technology advances to the point they are expendable after a single season/film/game and the company is sitting on hours of tape. And hope the point where technology advances to the point voice can be modulated from scratch happens past their careers.
posted by lmfsilva at 11:03 PM on November 5, 2016 [1 favorite]

So, this means I might finally get my sexy Kathleen Turner voice pack for Cortana?
posted by Samizdata at 11:34 PM on November 5, 2016

you can basically produce convincing footage of public figures doing and saying anything.

Although you currently have one public figure in particular who's in the habit of saying more ridiculous things than even the most mischievous user of this app could come up with.
posted by Grangousier at 4:06 AM on November 6, 2016

Better software (or a human) would have searched the available Keegan audio for a smoother transition from "my" to "wife" instead of using the existing "my" with the audible glide toward the "d" in "dogs". As it is, the first example sounds too much like "myd wife". A careful audio sculptor who understands pronunciation could pull the necessary transition from "I would" or "highway" or "why wait" (if the pitch and inflection matches nicely) to make that much more believable.
posted by scrowdid at 9:19 AM on November 6, 2016 [1 favorite]

.... And just like that, scrowdid ended up on a government watchlist.
posted by webmutant at 11:13 AM on November 6, 2016

Adobe would do well to get Tom Baker all over this.
posted by juiceCake at 1:08 PM on November 6, 2016 [1 favorite]

Fully expecting the bastard child of this and Autotune to make disposable commercial music even more so.
posted by flabdablet at 7:33 PM on November 6, 2016

So is this based on a sampled syllabary/corpus of the speaker's voice? It's hard to imagine that there's enough data in the waveforms (which usually don't show pitch/resonance but only amplitude) to get anywhere near close.
posted by yellowcandy at 10:15 AM on November 7, 2016

.... And just like that, scrowdid ended up on a government watchlist payroll.
posted by me & my monkey at 1:37 PM on November 7, 2016 [1 favorite]

juiceCake, do you mean Troy Baker? Nolan North? Jennifer Hale? The ones currently on strike?
posted by reiichiroh at 1:38 PM on November 7, 2016

Simpsons did it.

Max Headroom went there first in the season 1 finale of the US show: Blanks, aired 5 May 1987. Near the beginning of the episode, Network 23 R&D wunderkind Bryce Lynch is showing off his new editing suite, "Data Rescan". By the end of the episode, they use it to put words in a politician's mouth, so they can air a clip of him stating a position opposite of his actual belief, in order to show him that the public strongly supports the edited version instead of his own view.
posted by radwolf76 at 4:31 PM on November 7, 2016 [2 favorites]

This is... a little obviously-fake in the example, but I'm assuming there must be some intonation and inflection controls they're not playing with. It would t take much tweaking of that to make it sound seamless.

Well, it's worth noting, that this technology is literally in its 1.0 phase. Yes, it isn't seamless at the moment, but assuming there is a commercial market that will spur future advances, it's safe to assume that this will get real spooky in no time.

Also, the bland assurances about how they are working on watermarking (which got a significant audience applause) to detect this aren't at all convincing to me - show me a watermarking technology and I'll show you someone that can remove it (and create software to programmatically remove it).
posted by el io at 5:01 PM on November 7, 2016

juiceCake, do you mean Troy Baker? Nolan North? Jennifer Hale? The ones currently on strike?

I don't know who any of those individual are. Tom Baker played the 4th Doctor in the original Doctor Who series and also famously did some voice work for BT:

During the first three months of 2006, his voice was used by BT for spoken delivery of text messages to landline phones. He recorded 11,593 phrases, containing every sound in the English language, for use by the text-to-speech service.

A sample.
posted by juiceCake at 11:50 AM on November 15, 2016