I'm from Britain and it rated my accent as 99-100% British the few times that I tried it. So not bad from my point of view.
Meanwhile, I wonder what Cambridge Consultants are going to do with a dataset of thousands of Hacker News users reading out the sentence "Please call Stella and ask her to meet Bob the frog at the store with three small red plastic bags".
Indeed. They took a bunch of smaller words and spliced them together to fake his voice print. Cambridge Consultants are clearly harvesting our voice prints for the CIA. ;)
Probably just make refinements to the model they're using to classify samples.
The sentence is obviously taken (with some modifications) from the sample used by the speech accent archive ( http://accent.gmu.edu/howto.php ), which was constructed specifically to include every relevant phonological context in American English. (Not that this was necessarily achieved, but it was the goal.)
I'm American and I get mixed results between 89% American and 53% British on my normal accent. My pronunciation of "call" and "to" seems to tip the scales towards British and most of my other words are not strongly highlighted.
I can reliably fake a British accent although I seem to get stuck on frog and store still sounding American. When I do the stereotypical Texas style "yee haw" accent I get 99% American.
It's actually a famous sample; I remember seeing it in linguistics classes. It just happens to contain a lot of words that reveal differences between English dialects.
Meanwhile, I wonder what Cambridge Consultants are going to do with a dataset of thousands of Hacker News users reading out the sentence "Please call Stella and ask her to meet Bob the frog at the store with three small red plastic bags".