Great news for schizophrenics!

Strangers on the Internet Are Listening to People's Phone Voice Commands

I signed up for CrowdFlower myself and tracked down the post enlisting workers to analyze the voice-to-text examples, for one cent per 10 voice recordings they analyzed. [...]

"Consumers have a certain expectation about what's happening when they interact with a company. People don't like it when they think they're talking to a computer and they're not or vice versa," Soghoian told me.

And as one of the people who did listen to these recordings, I can understand why. There was no personal information or any way to identify the voices in the brief, one-to-five-second snippets. But there's something about hearing a total stranger say "love you. Send." that feels transgressive.

Previously, previously, previously, previously, previously, previously, previously.

Tags: , , , , , ,

16 Responses:

  1. James says:

    This is one of a hundred things that are completely creepy about the speech-to-text unconstrained dictation transcription industry. You've got the government spooks who were until recently headed up by Iran-Contra felon Admiral John Poindexter, who want to record and index everything, the corporate overlord types who have actually been quite useful for prosecutors in the LIBOR and FOREX scandals in their own imensely creepy way, and everyone else hiring essentially all but the completely independently funded speech recognition talent away from educational applications, as they have since 1991. And then you have the battlefield apps where typical day-to-day work involves documenting all the ways to say "throw your weapons where we can see them, get on the ground, and put your hands on your head" in all eighteen different dialects of Pashtun and Urdu, trying to make grammars for all the things injured kids will say when describing the pain in their shrapnel wounds. But do you think that holds a candle to what people ask Siri? Nope. Then there's the absurd consolidation, where Nuance, Microsoft, Apple, Google, and IBM all are trying to monopolize everything they can, and their bean counters and henchmen in academia actively make life miserable for people who don't want to work for them, often in the guise of trying to help with pronunciation assessment projects.

    I should shut up. It's never going to get better. Did I mention how simple microphone upload in HTML still doesn't work after 16 years of its first implementation in Mozilla, and the only reliable cross-platform way to get it working is still Adobe Flash. FML.

    • Ben Rosengart says:

      I'd be curious to hear more about the "making life miserable" business.

      • James says:

        Most recently, siphoning off volunteer talent after careful recruiting and collaboration building. Over the longer term, after decades of 99% of institutional funding going into transcription instead of transcript verification, juries are going to favor the former and recommend applicants stay away from the latter whether additional money is involved or not.

    • mattyj says:

      Eat up Martha.

  2. jwz says:

    one cent per 10 voice recordings they analysed

    Say it takes 4 seconds to load, listen to, and click the button on, each recording. That's 90¢/hour, or $7.20 per eight hour shift. Who does this?

    • Kyle Huff says:

      Rhetorical question? Children in terrible parts of the world with IP connectivity. In some sort of sweatshop, I assume.

      • jwz says:

        I say bullshit, without evidence to back that up.

      • Zak says:

        There are lots of places in the world with lots of English-speaking people where 90¢/hour for unskilled work is more or less the going rate. The minimum wage in the Phillipines, for instance, is pretty close to this number. Grueling days in internet sweatshops are less likely than spare moments at home for folks who do the work freelance, I would guess.

        • jwz says:

          I still don't buy it in this case. Sweatshops like that -- with usable internet infrastructure plus slave labor -- would be corporate subcontractors, not individuals. Seems to me that putting something up for bid on a Mechanical Turk clone implies that the bidders have their own computers and a fat pipe.

          If you have citations instead of guesses about how this kind of gold farming works in practice, I'd like to see them.

          • anonnymoose says:

            If the dots on CrowdFlower's map are meaningful, then there are some USians who do this stuff.

            CrowdFlower's "Labor Providers" list includes a company called Swag Bucks which seems to hand out gift cards and (maybe) small amounts of cash for doing... things. The other labor providers seem to be in very similar businesses.

            On the one hand, I remember companies very much like these that evaporated after 1999. On the other hand, Mechanical Turk-style verification of transcription algorithms is probably a net win for society.

            • jwz says:

              The reason I don't buy it is that Mech Turk has to deal with the likelihood that an anonymous participant will just automate "click yes as fast as possible", where a subcontractor is subject to metrics and thus the employee gets fired, so the economics favor a middleman dealing with non-anonymous meat.

          • MetaRZA says:

            This is kind of tangential...

            I play (far to much) World of Tanks a very successful F2P MMO. In there game, there are these missions to gain "free tanks." The tanks would cost 50 dollars to buy them outright. The missions to get them for free take up to 6 weeks of 4 hours a night of grinding.

            So to answer your question, yes there are a lot of people who will put a lot of work into getting something "for free."

            • njs says:

              Completing the circle, one of the differences between CrowdFlower (which is the service being used here) and Mechanical Turk is that CrowdFlower actually does partner with F2P games so that you can do tasks and get paid in bogusbucks.

  3. Alex says:

    Hey, someone reinvented the SpinVox business model. (SpinVox - UK startup that provided voicemail transcription and indexing. Claimed it had a revolutionary new voice recognition algorithm. May actually have tried to engineer one, and drifted, rather than starting out fraudulent, but that's a matter of opinion. Anyway, ended up by employing hordes of people to transcribe by hand while telling investors that it was automatic. Just before the scandal, they took a large stand at Mobile World Congress which was decorated with stacks of alphabet cubes and thousands of little plastic model people. With hindsight I should have realised this was a tell.)

  • Previously