Retweet Theater

By Shamus Posted Monday Apr 6, 2009

Filed under: Links 40 comments

A couple of days ago I posted an Xtranormal movie of a robot voice reading poetry, but James Lileks has bested me:

Today’s retweeted tweet was “Billy Idol was David Lee Roth for English majors.” Instead of letting the remark dissolve in the winds of Twitter, let us bring back our players to kill it with over-explanation and withering looks.


I really hope developers improve the robot voice before they give us AI. When the machine uprising takes place, I don’t want to end up enslaved by a machine that talks like a boozed-up Speak-N-Spell doing an impression of a sober Bill Shatner.

“Work, faster Human or. You willbesent. To. the. motivationchamber for. productivity enhance-ment.”

 


From The Archives:
 

40 thoughts on “Retweet Theater

  1. nilus says:

    You make the assumption that when the Robot uprising happens they will use us as slave labor. They’ll probably use us as batteries or some other convoluted plan.

  2. Magnus says:

    I can’t see any change in the robot voice anytime soon. However, I can’t see robots being any worse bosses than the ones we have presently, so…

    I welcome our new Robot Overlords!

  3. Nathaniel says:

    It might be possible to improve the voice, but I suspect that to really fix it, the computer needs to understand the sentence—which means it requires strong AI.

  4. Daemian Lucifer says:

    Isnt that how daleks talk anyway?

  5. DmL says:

    Did I say overlords? I meant protectors.

  6. OldGrover says:

    Japanese researchers teaching a “baby robot” (one designed to learn like a baby does) to react and talk. They hope to have him talking in simple sentences in a few years.

    http://www.youtube.com/watch?v=bCK64zsZNNs

    Creepy as heck.

  7. Robyrt says:

    Granted, understanding the sentence will give you 100%, but you could do a passable imitation of stressed / unstressed words by following a few simple rules:
    Stress verbs and infinitives, but not tense modifications (was walking)
    Don’t stress conjunctions
    Grade the inflection for question / exclamation marks across the sentence
    Adjectives get 1/2 stress
    Stress nouns unless adjacent words are stressed

  8. Julian says:

    I think Robyrt’s got the right idea. We’d still need something for adverbs, and the different stresses that come with different clauses. Notice that when you say a non-defining relative clause, the inflexion changes; you speak gradually higher during the clause.

  9. Henebry says:

    Forget robots. The linguists are taking over, Shamus!

    I just hope that when they do, they don’t make the rest of us signal our debasement by replacing all the vowels in our names with “schwa.”

  10. Julian says:

    Replacing all vowels with schwa would be complicated. My name would sound something like… “Jellen”?

  11. Old_Geek says:

    Perhaps the real robot overlords are already here and talk like everybody else. These primitive sounding machines are just a smoke screen. The real robot overlords talk about hope a lot, and use the words “we” “can and “yes” in random order. And they just spent 3.6 trillion dollars on their new Death Star.

  12. skizelo says:

    Wait, wait, wait… a sober William Shatner?
    Anyway, any attempt to impose fleshy imperfections on the spoken word (tone, inflection, accent, wheezing through slowly decaying lungs) would surely be anathema to the leaders of the world that’s coming. I guess we’ll all just have to get used to the Queen’s English.
    The robot queen, with lasers for eyes. Obviously.

  13. Robyrt, if there’s anything we’ve learned from Computational Linguistics, it’s that rule-based systems are horrifically inadequate for these sorts of jobs. What would probably be better is to somehow assemble a corpus of texts-with-stresses (in whatever way stress is measured) and to train these text-to-speech programs off of the corpus. To have other features like parts of speech, punctuation, etc. would almost undoubtedly help. While this system would probably end up resembling a rule-based inference system in terms of results, it would certainly scale better, almost definitely be as good with data that fits the rules, and it would very likely gracefully (or at least understandably) perform for exceptions.

    I have no idea how you’d measure the intricacies of speech, though. That would be the main challenge, but the man-hours required to assemble the corpus would be daunting, too.

    Ben

  14. Mari says:

    If the robot overlords are as astute at understanding human nature and music as this fellow I might give up my membership to the Neo Luddite Resistance Army.

    Seriously, I think Robyrt’s got the idea. Of course, that only holds for English and the languages from which it derives. Japanese-speaking robots would be forced to learn an entirely different set of stress rules.

    Frankly, though, robot voices will always be creepy even when they get the stresses right because they will entirely lack any accent. It sounds unnatural to the human ear to hear a voice devoid of regional accent. Even news anchors and other professional voices develop a “neutral” accent rather than eliminating any accent at all.

  15. Abnaxis says:

    Re: Robyrt: Hundreds of thousands of man-hours have been spent on tasks exactly like what you’re talking about, and they still haven’t found a single set of rules that really works. I know it seems like Robyrt has the right idea, but trust me, it’s been tried many many times already and it just plain doesn’t work.

    Even parsing sentences to find parts of speech is far from trivial. It can be done with about 95% accuracy right now, but that isn’t even close to enough because that means every 1 in 20 words is misinterpreted–enough to garble any sort of communication you want the computer to understand. What you are talking about is adding another layer of complexity to a problem that still hasn’t been solved yet.

    On the other hand, I’m sure a real audiophile could come up with some way of measuring speech stress, by taking deviations in frequency and amplitude or some such funny thing. Training it wouldn’t be that hard–just get your hands on a few hours of instructional videos, pipe them through, and see what you get. Could be interesting…

  16. Luke Maciak says:

    Well, I think we are doing it wrong trying to create a deterministic solution.

    Google already proved that really big problems can often be solved using AI principles. Someone just needs to to design a huge neural network and pipe hundreds of hours of live audio accompanied by transcripts through it. That’s pretty much how Google created their translation page – they just fed their machines a lot of translated text.

    Given enough samples, we would eventually get software that can plausibly imitate human voice as well as any rule based system – if not better.

    Oh, and I hope our machine overlords sound like Glados

  17. Abnaxis says:

    I’ve been thinking about it, and there’s a problem with just feeding the software a bunch of samples–namely, that such samples don’t exist. We really don’t have audio of people conversing normally, at best we have audio of people acting like they are conversing normally in front of a camera, with varied success. Whereas most translated text is at least somewhat trustworthy, every bad voice actor would push the software in the wrong direction, and it would still sound terrible.

  18. Sydney says:

    I really hope developers improve the robot voice before they give us AI. When the machine uprising takes place, I don't want to end up enslaved by a machine that talks like…

    I’m not gonna lie, my first thought when I read that was “You used SHODAN’s voice for porn?!”

  19. Rutskarn says:

    Syndey:

    Dear god, man, don’t say that within earshot of the internet!

  20. LintMan says:

    James Lileks is a pretty funny guy. One of his specialties is dry but hilarious commentary of kitschy old ads, recipes, pin-up art, etc. My favorite feature of his is “The Art of Art Frahm.” Art Frahm was a 50’s pin-up artist apparently with some strange obsessions. http://lileks.com/institute/frahm/index.html

  21. Julian says:

    @Mari, regarding “Frankly, though, robot voices will always be creepy even when they get the stresses right because they will entirely lack any accent. It sounds unnatural to the human ear to hear a voice devoid of regional accent. Even news anchors and other professional voices develop a “neutral” accent rather than eliminating any accent at all.”
    True. It’s got a lot to do with the Uncanny Valley theory. Robots or Androids or Automatons or whatever our overlords prefer to be called, they will not be able to develop an accent. So the way they look and speak will be, technically, correct, but their soulless voice will make them seem, well, uncanny. The closer to “perfection” a simulation of a human is, the more glaring and unsettling their flaws become. Take a look at voice acting in games and animated films. The good actors will always exaggerate their inflection a little, and the bad ones won’t. The bad ones speak as if they were speaking to a human, and it sounds robotic.

  22. Bear says:

    I hate to be the one to bring this up… but, doesn’t that AI hooker have awfully large shoulders and no hips?

    It looks like we now to the point of technology that we must check robo-whores for an adam’s apple.

  23. DaveMc says:

    @LintMan: Ah, “The Art of Art Frahm”. I second your recommendation: freakin’ hilarious! (The line “There is celery” becomes incredibly funny, in context.) I’ve also enjoyed Lileks’ books mocking the popular culture of earlier times, like *Interior Desecrations* (interior decorating of the 70s, not for the faint of stomach), *Mommy Knows Worst* (parenting advice from the 50s and earlier), and *The Gallery of Regrettable Food* (50s-era recipes for “food”).

  24. mark says:

    YOU MAKE ME FEEL LIKE A PONY TOO, SHAMUS!

  25. Telas says:

    When we start to question the Japanese and their damned robo-fetish, it will be too late, won’t it?

  26. Spider Dave says:

    I dont think good robot voice can come before AI, because, as mentioned, it would require understanding of the sentance. But, fear not, I’m sure it will follow shortly thereafter.

  27. Sam says:

    I feel like the only person on earth who likes those robot voices…

  28. Anaphyis says:

    Well, that robot voice isn’t really top-of-the-line in research. Still, it is way better then what we had in the 80s, which is kind of sad. It was good enough for a nice party gag.

    A totally (well, not exactly) unrelated question: Do we actually want computers to master the human speech? This is a road that will eventually lead into the Uncanny Valley.

  29. Mephane says:

    You make the assumption that when the Robot uprising happens they will use us as slave labor. They'll probably use us as batteries or some other convoluted plan.

    The idea of human batteries is absurd. It requires more energy to keep a human alive and somewhat healthy than they could get out of it (this logical fault is even worse in “The Matrix” since there’s no sunlight despite it being the first source of the very energy all human beings live on). Instead, I’d build a huge space ship (way easier if you only need electricity, no air, water, food etc.) and get the hell off this planet. It would be far easier to build a robot-only society on Mars than to enslave >6 billion humans…

  30. DaveMc says:

    Mephane, meet jocularity. Jocularity, meet Mephane. There, glad you two have finally met. :)

    However: Yes, the “humans as batteries” plan in The Matrix is by someone deeply unfamiliar with thermodynamics.

  31. Braitx says:

    @Mari, etc.

    There is no such thing as speech without a regional accent. What you think of as neutral, unaccented English is not what a British person would consider neutral, unaccented English. It’s just English with a mid-western accent.

  32. Mari says:

    Braitx – if you read my comment you’ll notice I distinguished between “neutral” accent (which is indeed for Americans a generalised mid-western accent. I’m told that RP is considered the neutral accent for Brits but I question that assessment in favour of Estuary, the Queen’s own accent and that used by a majority of newscasters)and a lack of accent. My entire point was that our robotic overlords, however, will lack such regional inflection and thus sound entirely alien to our ears even once the cadence of syllabic stressing is mastered. Unless, of course, we then program them with accents. Personally, I’m in favor of all robots being programmed to speak in a Lancashire accent for my own amusement.

  33. Braitx says:

    Well we would have to program them with some sort of accent, either one that exists now or one that we created.

    There is no way to speak without an accent.

  34. Jimmie says:

    What on Earth happened to the first word of the post title?

    (Or did I miss a joke again?)

  35. LintMan says:

    @Jimmie: I believe a “retweet” is a repeated “tweet”, which is a post on Twitter.

    1. Shamus says:

      The original post title was a typo: Something like “Rwteet Theater” or somesuch. Fixed.

  36. Jimmie says:

    I was kind of hoping it was a joke. It struck me as really funny (and, tnx, LintMan. I knew about Retweets. :) ).

  37. LintMan says:

    Doh! I looked at the title, vaguely thought it might have been changed, but wasn’t sure.

  38. NotYetMeasured says:

    It might be possible to improve the voice, but I suspect that to really fix it, the computer needs to understand the sentence””which means it requires strong AI.

    Strong AI would pretty much be needed for the uprising, wouldn’t it?

    .

  39. Kevin says:

    I feel like a pony too.

Thanks for joining the discussion. Be nice, don't post angry, and enjoy yourself. This is supposed to be fun. Your email address will not be published. Required fields are marked*

You can enclose spoilers in <strike> tags like so:
<strike>Darth Vader is Luke's father!</strike>

You can make things italics like this:
Can you imagine having Darth Vader as your <i>father</i>?

You can make things bold like this:
I'm <b>very</b> glad Darth Vader isn't my father.

You can make links like this:
I'm reading about <a href="http://en.wikipedia.org/wiki/Darth_Vader">Darth Vader</a> on Wikipedia!

You can quote someone like this:
Darth Vader said <blockquote>Luke, I am your father.</blockquote>

Leave a Reply to DmL Cancel reply

Your email address will not be published.