Mark Williamson wrote:
Now, that aside, there's another major problem: accent. Who's to say I could understand what you were saying, if you said "star"? There is no unambiguous way to represent words through audio.
That's a theoretical possible problem rather than actual data (which is why that' s what I asked for) or even anecdote.
I haven't found data yet, but here's a page of theory with anecdote:
http://www.standards-schmandards.com/index.php?2005/01/01/11-captcha
Note they get around your problem by using numbers. This apparently worked on three casual test subjects. Though that, of course, is anecdote, not data.
The W3 paper just provides possible approaches with no words on effectiveness: http://www.w3.org/TR/turingtest/
Evidently the audio option was frequently unusable a couple of years ago: http://news.com.com/2100-1032-1022814.html - I would *presume* there's been improvement since then.
Does anyone have or know of *actual data* (rather than hypothesis or anecdote) on whether non-visual captchas are any good as yet?
- d.
On 11/2/05, David Gerard dgerard@gmail.com wrote:
Does anyone have or know of *actual data* (rather than hypothesis or anecdote) on whether non-visual captchas are any good as yet?
Anecdote, again, but it is a success story: when we launched CAPTCHAs on LiveJournal which included audio support, we got a surprising response of thank-you letters from blind users (maybe two or three, which is pretty significant in comparison to the feedback about other features...).
One way to look at it is: with a visual CAPTCHA, a blind user is completely out of luck. So even with awful usability, it's better than nothing.
Evan Martin wrote:
On 11/2/05, David Gerard dgerard@gmail.com wrote:
Does anyone have or know of *actual data* (rather than hypothesis or anecdote) on whether non-visual captchas are any good as yet?
Anecdote, again, but it is a success story: when we launched CAPTCHAs on LiveJournal which included audio support, we got a surprising response of thank-you letters from blind users (maybe two or three, which is pretty significant in comparison to the feedback about other features...).
I was going to mention LiveJournal. :-) I have personally seen one of those "thank you" support requests from a blind user.
But I was going to add: David, you can try it yourself. Just go to http://www.livejournal.com/create.bml and take the audio test. Personally I actually found it easier to pass than the visual one - the letters in the visual captchas are sometimes warped enough to make them ambiguous even for humans (e.g. g vs. q).
Timwi
David Gerard wrote:
Does anyone have or know of *actual data* (rather than hypothesis or anecdote) on whether non-visual captchas are any good as yet?
To address one particular type: a conference paper reports experiments that suggest using speech synthesis degraded by noise is not an effective basis for audio CAPTCHA:
"...although there seems to be a gap in the ability of understanding synthesized speech with background noise between humans and computers, our results discourage using this gap to build an audio-based CAPTCHA"
Only the abstract is available online for free. The full paper would require an online purchase or a library visit.
Tsz-Yan Chan. "Using a Text-to-Speech Synthesizer to Generate a Reverse Turing Test," /ictai/, p. 226, 15th IEEE International Conference on Tools with Artificial Intelligence (ICTAI'03), 2003. **http://doi.ieeecomputersociety.org/10.1109/TAI.2003.1250195
Though the paper is from 2003, one would guess that particular CAPTCHA schemes get less, not more, effective over time as automated recognition technologies improve.
wikitech-l@lists.wikimedia.org