I'm curious: what method specifically are you referring to? I've seen talks on getting through Google's new NoCatpcha Recaptcha a small percentage of the time (which is worrying but nowhere near as troubling as the susceptibility to OCR most other captchas suffer from) but I wasn't aware of any reliable and automated way to bypass it (since outsourcing them to human captcha solvers via a paid (~2 bucks per 1k captchas) API isn't exactly automated).
I was being purposely vague. However, there's a method of automatically downloading the sound files when requesting for a audio version rather than the images to help with usually impaired people. Generally, people write a Python script, automatically download it, and then can input it. I haven't tried this myself, but have seen it working on a test site from a friend of mine. He kept on having bots sign up to his website out of nowhere, and he found that this was the method they were using. Basically though, you're coding something that will be able to convert the webpage so you can capture the javascript, and then you are automatically downloading the sound file, and then run it through a text to speech program or something along those lines. Its obviously not going to be 100% accurate, but definitely somewhat automated. We've looked into options to prevent this, but it doesn't seem possible as Google doesn't allow disabling the audio option. Unless, we are missing something.
Obviously, I haven't checked whether this can be done on Bitcointalk, because I haven't logged out in a few months. But, it shouldn't be any different to what I witnessed a few months ago.