Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Robotics AI

'Deep Fake' Technology Used to Perfectly Re-Create a Radio Announcer's Voice (reuters.com) 44

Posted by EditorDavid from the DJ-fake dept.
For 32 years a human named Andy Chanley has been a radio announcer (now working afternoon's at Southern California's 88.5 KCSN), Reuters reports. But now.... "I may be a robot, but I still love to rock," says the robot DJ named ANDY, derived from Artificial Neural Disk-JockeY, in Chanley's voice, during a demonstration for Reuters where the voice was hard to distinguish from a human disc jockey.

Our phones, speakers and rice cookers have been talking to us for years, but their voices have been robotic. Seattle-based AI startup WellSaid Labs says it has finessed the technology to create over 50 real human voice avatars like ANDY so far, where the producer just needs to type in text to create the narration....

Martín Ramírez, head of growth at WellSaid, said once the voice avatars are created, WellSaid manages the commercial agreements according to the voice owner's requests. WellSaid voice avatars are doing more than DJ work. They are used in corporate training material or even to read audiobooks, said Ramirez.
The article points out that while (human) announcer Andy Chanley was recording his voice, he discovered he has Stage 2 lymphoma. While he eventually recovered, Chanley liked knowing that there was also another way that the sound of his voice could still be supporting his family — and that his grandchildren could hear the sound of his voice.
This discussion has been archived. No new comments can be posted.

'Deep Fake' Technology Used to Perfectly Re-Create a Radio Announcer's Voice More

'Deep Fake' Technology Used to Perfectly Re-Create a Radio Announcer's Voice

Comments Filter:

  • Olde singers (Score:4, Insightful)

    by flyingfsck ( 986395 ) on Sunday December 05, 2021 @04:56AM (#62048709)
    So, pretty soon we will be treated to a zombie horde of long dead and faded pop stars releasing new numbers. There were a few of those in the 1990s, but those were remixes in duet with younger singers for example Sinatra with Dion.

    • Re: (Score:2)

      by DrXym ( 126579 )
      I doubt anybody would be interested in buying a fake Frank Sinatra track. After all there are any number of Sinatra impersonators out there who could already do it and more convincingly than any computer is likely to do any time soon.

      It might be useful pop stars who are just really bad at singing, the kind who already resort to the autotuner who might look upon this as a step up.

      • Re: (Score:2)

        by fazig ( 2909523 )
        Video game makers could be very interested good synthesized voices.
        And of course writers would likely be among the interested, to have their text read back to them in a decent voice, that way it's a lot easier to catch mistakes during editing, and of course it could also possibly be used to generate audio books.

        There's already services available like this: https://syntheticaudiobooks.co... [syntheticaudiobooks.com] though given the examples they use to advertise their service there, there's still a lot of room for improvement.

        • Re: (Score:2)

          by bn-7bc ( 909819 )
          Or audible for that metter, producing audiobook would be a bit cheaper when you don't need narrators I would think. ok the license for the tech will not be free but think of an almost instantaneous text to audiophile conversion and maybe even auto synced text highlighting when read.

          • Re: (Score:2)

            by fazig ( 2909523 )
            It could be considerably cheaper. And you could possibly give every character and narrators their own distinct voice (not the literary voice that every well written character ought to have already).

            Though I wonder how things like the nuances of tone of voice can be handled. Human narrators do that by comprehending the writing and context. Could a neural network learn that without requiring additional syntax?

      • I doubt anybody would be interested in buying a fake Frank Sinatra track. After all there are any number of Sinatra impersonators out there who could already do it and more convincingly than any computer is likely to do any time soon.

        It might be useful pop stars who are just really bad at singing, the kind who already resort to the autotuner who might look upon this as a step up.

        Agreed. If synthetically resurrected pop stars were to have appeal I'd expect Elton John impersonators [youtube.com] to be selling out stadiums and not doing random gigs.

        Only a minor part of being a music celebrity comes from the voice or even the technical talent. The majority comes from songwriting, composition, and personality (incl. marketing). The songwriting and composition die with the celebrity (and can come at least partially from 3rd parties), and the personality requires an authentic connection.

        If you've writt

      • Probably can be used in advertising and such.

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      I'd like to see old TV series revived. Proper endings where they didn't get one in the original run. Maybe even complete new seasons. With deepfake for faces and voices it's possible to do, even if the original actors are deceased.

      • Yeah, let's give The Sopranos that Dunkaccino ending [youtube.com] it always deserved.

        And let's finally see Casablanca in proper 4DX, with a new, surprise ending and a post-credits scene announcing a shared universe.

      • youtube.com/watch?v=Fcn4p213Zg8 - Creepy Orville Redenbacher Commercial

        This was done years ago, and though I don't see anything wrong with the commercial on the surface, there are enough subtle clues that let people know that something is "wrong", or "off" and it had creeped a lot of people out.

          Look up "Uncanny Valley".

          We should just let the dead rest.

    • My question is will people be interested in a synthetic "celebrity." That would be of great value to media companies because once everybody is used to somebody in a specific role, they are irreplaceable and demand millions of dollars. Not because they do the job "better" than everybody else, but because they are now the only person who can fulfill audience expectations.

  • Not perfectly (Score:5, Interesting)

    by phantomfive ( 622387 ) on Sunday December 05, 2021 @05:02AM (#62048721) Journal

    The headline contradicts the summary/article. The headline says "perfectly" the article says:

    "a demonstration for Reuters where the voice was hard to distinguish from a human DJ."

    So not perfect at all.

    • Here is Styx: https://www.youtube.com/watch?... [youtube.com]

    • The headline contradicts the summary/article. The headline says "perfectly" the article says:

      "a demonstration for Reuters where the voice was hard to distinguish from a human DJ."

      So not perfect at all.

      ... but plenty good enough to fool tens of millions of regular people who are not experts in audio deep fakes.

      • Re:Not perfectly (Score:4, Funny)

        by infolation ( 840436 ) on Sunday December 05, 2021 @10:38AM (#62049115)
        Sassy Justice asked Michael Caine and he said 'watch with your ears'. [youtube.com] Wise words for all of us.

        Fred Sassy:

        As a black belt in Taekwondo Self defense comes naturally to me.
        But how do we defend ourselves against Deep Fakes?
        My next guest says all we have to do is listen. It's Acadamy Award winning actor Michael Caine.

        Michael Caine:

        The human brain is a very clever thing. And it can detect the differences in what is your real voice And what is a fake voice.
        You know I saw the Tom Cruise deep fake and like everbody else I was like Blimey! That's Tom Cruise!
        But I listened. And you know what?
        There's little differences in the deep fake Tom's voice and real Tom's.
        And I listened And I could hear differences in the voice. Little tiny little mistakes.
        To tell a deep fake all you got to do is...
        Watch. With your ears.

  • This is something I want to read in a paper? (Score:4, Insightful)

    by c120plus ( 3825447 ) on Sunday December 05, 2021 @05:14AM (#62048735)

    It totally makes sense to report this in a printed paper, but on the web, why not include a sample audio/video file so that we can hear the robotic voice?

  • The link goes to a plaintext news story. Surely there is a better link that could have been used that actually demonstrated what was achieved with audio. Something that might also describe the process, how samples were used to train the simulator, the limits of the technology, the ethics of doing so etc.

  • Better but I could still tell it was generated (Score:3, Interesting)

    by slowmovingtarget76 ( 911033 ) on Sunday December 05, 2021 @07:43AM (#62048883)
    The clip they play where his AI-generated voice says heâ(TM)s a DJ is probably what they played to their investors. As soon as the engineer typed the sentence and played back the on-the-fly audio I could hear the imperfections. Itâ(TM)s certainly better than where we were just a few years ago and a ton better than those crappy ones you hear reading text on YouTube videos but itâ(TM)s nowhere close to perfect. I believe the guy when he says this improves constantly. I just think theyâ(TM)re going to need another year or two before itâ(TM)s ready for prime time.
  • Just like we can create perfect images of people that don't exist, it must be possible to create human voices that don't exist. Why copy one particular voice?

    • Re: (Score:1)

      by Anonymous Coward

      Just like we can create perfect images of people that don't exist, it must be possible to create human voices that don't exist. Why copy one particular voice?

      To fake, say, Ilhan Omar's voice trash talking Jesus so your fellow nutbags have a reason to break out their arsenal and 2nd amendment her? Or to fake Donald Trump's voice saying something that gets him in hot water with swing voters? The potential benefits of creating an AI that can deep fake the voice of a particular person are endless.

    • Re:Why? (Score:4, Insightful)

      by Jarik C-Bol ( 894741 ) on Sunday December 05, 2021 @09:34AM (#62049025)
      Practice. To make a perfect original *anything* you practice making perfect copies of the same type of thing, until your skill is high enough that there is no discernible difference between yours and the originals, then you make a new unique one with these mastered skills.

  • My voice is no longer my passport... (Score:3)

    by denzacar ( 181829 ) on Sunday December 05, 2021 @09:42AM (#62049043) Journal

    I guess rectal probe is the only biometric still valid.

    Checking email and starting thermonuclear war is sure going to be a pain in the ass from now on.

  • I for one welcome this. I'm looking forward to "Walter White" announcing that the pressure cooker is done cooking or "Jack Torrance" letting me know there's someone at the door.

    • Re: (Score:2)

      by bn-7bc ( 909819 )
      Nah it has to be Jim Carter as Charles Carson doing the doorbell announcements nothing else will do, well maby the old footman if Carson is down for the day :)

  • "I'm the very best bot and I will live forever and ever, everyone knows it! I'm not going away, you low-life orange haters! I'll be in every flying car and more popular than that ugly trash-can R2D2. The "2" is because he's a two. Should really be R0D0."

    "I can break-dance and blow bubbles for the kids at the same time and millions will clap! When you need a lout for parties, I have a pussy grabber attachment. I grabbed Siri and Alexa already; they loved it, despite what lyin' CNN said. I'm a grabbing genius

  • I just tried the demo (Score:3)

    by presearch ( 214913 ) on Sunday December 05, 2021 @05:03PM (#62050005)

    It's remarkably good, especially at pronouncing proper names correctly.
    The voices that they let you try are a little sales-weasel oriented,
    but I couldn't hear anything synthetic in the paragraphs I tested with.
    It sounded like a pro voice-over actor.

    It's obvious that Morgan Freemans etc. will be forthcoming,
    and indistinguishable from the real person. Entertainment lawyers
    will have plenty of work.

    I don't think the product is worth what they want to charge every month,
    but it shows how good the state of the art can be.
    Soon, the synths that come with the OS will all be this good.

  • I can see it already, big studios won't sign recording contracts with new talent unless they give up rights to their voice after their death.

Slashdot Top Deals

HELP!!!! I'm being held prisoner in /usr/games/lib!

Close