Developers at business AI company Dessa have come up with a new text-to-speech system called "RealTalk". In the version they demoed, it was trained to speak with the voice of popular podcaster Joe Rogan. The developers have put up a site with a blind test at http://fakejoerogan.com/. They must have been so impressed by their own creation that they discuss the implications at https://medium.com/@dessa_/real-talk-speech-synthesis-5dd0897eef7f.
Your humble submitter did the blind test and just barely had a majority of correct guesses, but was so impressed by the quality that he considered it newsworthy - how do you fare in the test?
Related Stories
Meta says its new speech-generating AI tool is too dangerous to release:
Meta has unveiled a new AI tool, dubbed 'Voicebox', which it claims represents a breakthrough in AI-powered speech generation. However, the company won't be unleashing it on the public just yet - because doing so could be disastrous.
Voicebox is currently able to produce audio clips of speech in six languages (all of which are European of origin), and - according to a blog post from Meta - is the first AI model of its kind capable of completing tasks beyond what it was 'specifically trained to accomplish'. Meta claims that Voicebox handily outperforms competing speech-generation AIs in virtually every area.
So what exactly is it capable of? Well, for starters, it can spew out reasonably accurate text-to-speech replications of a person's voice using a sample audio file as short as two seconds, a seemingly innocuous ability that holds a huge amount of destructive potential in the wrong hands.
[...] Meta clearly believes its new tool is good enough to fool at least the majority of people [since] it's explicitly not releasing Voicebox to the public, but instead publishing a research paper and detailing a classifier tool that can identify Voicebox-generated speech from real human speech. Meta describes the classifier as "highly effective" - though notably not perfectly effective.
[...] A little caution, patience, and respect for the magnitude of this technology is a welcome sight - although I doubt Meta will sit on Voicebox for too long, since the shareholders will no doubt be wondering how much money it can make them...
(Score: 2) by takyon on Monday May 20 2019, @02:25PM (2 children)
It seems to blow the tinny Lyrebird demos out of the water:
https://www.theverge.com/2017/4/24/15406882/ai-voice-synthesis-copy-human-speech-lyrebird [theverge.com]
But we can't be too sure, because:
This is the same copout crap [soylentnews.org] we see from OpenAI. In this case, it's easy to lecture people about ethics, when releasing the research would help your competitors.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 3, Insightful) by Rich on Monday May 20 2019, @03:10PM
I had a similar feeling about that non-release, but I'm fairly certain that any party able to do a billion-dollar buyout already has an ethics panel that ensures the knowledge rests in very good hands. :)
As for how it works, the former papers of their mastermind Alex Krizhevsky and the traces the other mentioned guys left behind on the net might be a hint. If I had to have a go at it, I'd model the vocal tract up to a point where phonetic decomposition and resynthesis of original voice is somewhat better than good enough, have a net that translates a sliding window of letters and mixed in grammar information from further away to phonemes. The grammar analysis would be required to get intonation and accents right, and resolve ambiguities; i'd try that with a separate net. But what net topologies to use, and how deep, i'm completely clueless. The guys involved have enough experience to choose this for the probability of something thrown at it to stick - and in this case it seriously did.
(Score: 4, Insightful) by RamiK on Monday May 20 2019, @04:53PM
Since when turning "Power to the People" to "Power to the Corporations" has become ethical? Have we learned nothing from all those DNA labs that eagerly falsified tests for the police? If you have technology that allows falsifying text / voice / video recordings and you decide not to release it, you're guaranteeing someone else will develop it and use it for evil. The only moral thing to do is release it and make it trivial to falsify those sorts of recordings so their creditably falls and society adapts by implementing secure encryption right off the sensors.
compiling...
(Score: 4, Interesting) by danmars on Monday May 20 2019, @02:54PM (3 children)
I got them all right first try. What is so hard about this? The AI version is terrible at determining when to pause or change pace or inflection. I imagine it would be even easier for someone who actually listens to Joe Rogan.
(Score: 3, Interesting) by Rich on Monday May 20 2019, @03:26PM
It was late at night when I did the test, and I was paying attention to local modulation glitches, rather than overall flow. I've been developing an electronic music instrument over the last years and am probably conditioned on picking up the slightest flaw in transients, rather than processing the composition of longer spoken text. I re-listened, focusing on pausing, like you suggested, and got 6 out of 8 right, one more than in my first listening - it is not as easy as you suggest, at least not to me (fluent, but non-native English, mostly exposed to German, test on MacBook Pro 2009 built-in speakers, dent at 3kHz in my right ear).
(Score: 3, Interesting) by Alfred on Monday May 20 2019, @07:29PM
I think they are getting closer to what the CIA/NSA was doing last decade. Part of that is to take a mic feed and translate it into different voice in real time. That analysis helps with landing the right inflections and stresses too.
(Score: 3, Interesting) by KilroySmith on Monday May 20 2019, @07:54PM
7 out of 8 for me - only missed the first one. If I had listened to several before guessing, then it would have been 8 of 8. The pacing is wrong, and there are fewer intonation changes in the faux Rogan selections, making it easy for me to tell the difference.
That said, they samples were remarkably good. I'm quite impressed.
(Score: 2, Funny) by Anonymous Coward on Monday May 20 2019, @09:33PM (1 child)
There is no proof that Joe Rogan isn't already artificially intelligent.
(Score: 0) by Anonymous Coward on Tuesday May 21 2019, @07:00PM
joe_rogan_mind_blown.jpg