The AI-powered unCAPTCHA system designed by University of Maryland reseachers can now defeat Google’s latest audio-based reCAPTCHA challenges with 90% accuracy. Simply put, the unCAPTCHA system downloads the audio challenge, splits it into several clips, then runs them through several text-to-speech systems to determine exact and near-homophones, weighing the aggregate results by confidence level before sending the most probable answer back to Google. Read more to see the first version in-action.
Previously, reCAPTCHA used digits, as in the recorded voice would say a series of numbers, but Google updated their systems this year, changing the audio challenges so that they use words rather than numbers, and improving browser-based automation detection. So, the UMD team also updated unCAPTCHA to be able to defeat this by using the same text-to-speech approach, and this time, unCAPTCHA2 uses a screen clicker to move to certain pixels on the screen to trick it into thinking a real human is interacting with the browser.