You should consider building your voice recognition system into captcha inputs that people could embed into their own webpages.  It would still be a visual captcha, since the person would have to read the word/letters - but you would provide the option of submitting it verbally to further the cause of open speech recognition.  Would obviouslly have to be on words that were good enough to give a fair probability score when they're being explicitly looked for.

Basically just another route for voxforge submission, but could help spread your cause.

--- (Edited on 4/23/2009 9:46 pm [GMT-0500] by Tebbra) ---