1linklistFREE TRIAL Linklists - VPM of 150+ - http://1linklist.com
edited March 2017
Very interesting. We actually played with something similar awhile back, before we gave up on making our own Audio OCR.
The problem I ran into came down to proxies; the audio becomes insanely garbled, to the point that even a real person cant understand it, very very quickly.
Basically Google seems to be overly vigilant with garbling the audio, and you would need a massive amount of clean proxies to avoid that filtering.
Will definitely play with this guys library though, he mentions just that problem in his post. Maybe he found a way around it?
There are image recognizing neural networks around that identify things in images. You could for normal captcha use the same principal and send all the images in a captcha to one of these image recognizing neural networks to identify the things in the objects and scan the text question to find what recpatcha is looking for???
Comments
The problem I ran into came down to proxies; the audio becomes insanely garbled, to the point that even a real person cant understand it, very very quickly.
Basically Google seems to be overly vigilant with garbling the audio, and you would need a massive amount of clean proxies to avoid that filtering.
Will definitely play with this guys library though, he mentions just that problem in his post. Maybe he found a way around it?
http://www.blazingseollc.com/ocr/api.php
you just always refused to have GSA grab the audio captcha as an option