Re: reCAPTCHA implementation problems

what follows is an excerpt from my reply to ben when i emailed him about 
the status of the plans he had outlined, and which can be found at:

http://lists.w3.org/Archives/Public/wai-xtech/2007May/0048.html

this was just a quick knee-jerk reaction to some of the problems ben 
is experiencing as an implementor so as to accessify and increase the 
usability of the reCAPTCHA interface.   for those of you joining this 
thread late, you can test the reCAPTCHA yourself at:

http://www.recaptcha.net/

basically what follows is some very basic low-level advice, but since i 
placed an excerpt from ben's personal reply to the wai-xtech list

http://lists.w3.org/Archives/Public/wai-xtech/2007Jul/0009.html

it is only fitting and proper to also archive the other side of the 
conversation, in an attempt to avoid redundancy at such a meta-level; 
i don't believe there to be anything inflamatory about the information i 
provided him, which is a polite way of saying, if you know of a screen 
reader developer (windows, linux, and mac) please post that info to this
thread, rather than argue over market-share for products x y and z

---- Begin Excerpt of GJR's Reply to Ben M -------
as for screen-readers, you can either try the free Non-Visual 
Desktop Access (NVDA) screen reader for windows: 

http://en.wikipedia.org/wiki/NonVisual_Desktop_Access (description) 
http://www.nvda-project.org/ (the NVDA project's site) 

there is also charles chen's FireVox self-voicing extention for 
FireFox: 
http://clc4tts.clcworld.net/clc-firevox_doc.html 

which is also free...  also, you might try contacting the major 
screen reader and screen magnification software developers -- i'm 
sure that there would be buy in for the screen-reader or screen 
readers that helped make reCAPTCHA accessible...  here are the 3 
major players in the magnification and/or screen-reader business: 

http://www.freedomsci.com/ (JAWS and MAGIC) 
http://www.gw-micro.com/ (Window-Eyes) 
http://www.yourdolphin.com/ (HAL and SuperNova) 

of particular note is that HAL offers the user a choice of 
keyboard layouts -- one can overlay the JAWS keyboard or the 
Window-Eyes keyboard (or use the native HAL keyboard) which 
makes the learning curve a little less steep, as it is easier 
to switch to something an individual user is familiar with -- 
the keyboard overlay for another screen reader's commands, but 
with the functionality of HAL or SuperNova (more bang for your
buck if you have to purchase)

i like the 7 out of 8 rule, but worry that those with cognative problems 
may be flummoxed by the backwards looping; i think a switch from the 
google-like implementation to alternating voices of different 
characteristics (male to female to deep male voice to childlike voice, 
etc.) would be sufficient to trick voice-recognition software, such 
as in the (real life) case of a blind user who also happens to be a 
double-amputee, and has to rely on speech-input as well as speech 
output: if the user is male, he needs to train the voice-recognition 
software to correctly interpret his vocal commands (while wearing an 
earpiece, so that the software cannot hear his screen-reader's output) 
AND then switch his screen reader voice and/or voices to the opposite 
sex, so that when both voice-input and voice-output are used together, 
it responds only to his voice, and NOT that which issues from a speech 
synthesizer, so that the voice-recognition software does not 
misinterpret synthesized speech, as issuing commands or entering text 
into an editable document; obviously, if the user is female (real life 
case: paralysis of one arm, and an arthritis in the other arm which is 
so severe that her hand is locked into a clenched position) you simply 
do the opposite of what i just outlined above -- use male voices for 
speech output, and train the voice-recognition software to recognize 
your (female) voice...

so, if each alpha-numeric character was spoken in alternating voices, 
such as "average male" to "average female" to "male child" to "female 
child", and so on, i think you could foil attempts at speech recognition
not by adding noise and backwards loops, but simply by changing 
voices within an audio captcha

thanks again for all of your time and patience (we are trying to 
get someone to take responsibility for firing aural clues using 
ACSS when triggered by user interaction or by their assisstive 
technology (especially curtis, as FireVox supports the 
CSS3-speech module) -- one path i'm ivestigating is to integrate 
XML Events into HTMLx (the work of the HTML WG) and into XHTML2 
precisely for the purpose of firing purely aural cues (which help 
a lot more people than the blind, when actually implemented) -- 
the problem is, no one wants to allow anyone access to their OS' 
native embedded audio renderer -- the beneath the hood engine 
that fires system sounds and plays embedded files without opening 
a media player instance, so the long-term approach is to define an 
XML Event that is interoperable and standard across platforms, so 
that there is more buy-in by developers...  but for now, let's deal 
with today's technological realities, with an eye to the future 
development path that the HTML WG ends up pursuing in the area 
of forms -- and, hopefully, that path will lead to integration with 
XForms...

type at you soon, 
gregory. 

--------------------------------------------------------------
DISCUSSION, n.  A method of confirming others in their errors.
                     -- Ambrose Bierce, The Devil's Dictionary
--------------------------------------------------------------
           Gregory J. Rosmaita, oedipus@hicom.net
       Camera Obscura: http://www.hicom.net/~oedipus/
--------------------------------------------------------------

Received on Wednesday, 11 July 2007 19:43:12 UTC