Normally, we translate effortlessly between accents, so much so that we don’t even notice that we’re doing it most of the time. If someone tells you their name and they have an accent different from yours, you repeat it back in your own accent, not in an imitation of the way they said it. Let’s say you’re from London and your friend is from Vancouver, and her name is Martina. She’s probably got a ‘rhotic’ accent so she’ll pronounce the ‘r’ in her name, but you probably won’t. When you say her name back to her to check you heard it right, you aren’t going to pronounce the ‘r’ just because she does.
That’s why it’s weird when this doesn’t happen. When Britney said ‘granny panties’, it wasn’t a phrase Frank had ever heard before. Britney’s accent is also very different from Frank’s, and when she said it, what he heard (and repeated back to her) was ‘grainy painies’. If you can’t relate the sound string you hear to a known word or phrase, the only thing you can do is approximate the way it sounded. What you say sounds just like the phrase but you don’t know what it is you’re saying.
I see something like this in my first year seminars. One week, we do an exercise where they have to work out what phrase is written in phonetic transcription. The way to do this is to ‘sound it out’. Often, they are literally saying the exact phrase perfectly, but they can’t hear the words or extract the meaning from the string of sound. It’s fascinating and completely hilarious.