In the broadest terms, I generally don't pronounce my code at all. I just write it.

If something has a real name, I tend to call it by that name and I choose names accordingly (e.g., singular nouns for scalars, plural nouns for arrays/hashes). In cases where it may be ambiguous, I'll pronounce the sigil as the normal character, with the one oddity that, unless I'm paying attention, I have a tendency to pronounce $ as "hex" instead of "dollar". (My first programming experiences were with Apple ][s, where hexadecimal numbers were writen as "$E8".)

For punctuation variables, there's not much choice but to read off the string of characters literally, although, as another oddity, I generally pronounce _ as "sub", making $_ "hexub" ("hex-sub"). Fortunately, I speak in Perl rarely enough that I've never yet encountered a situation where it was ambiguous whether "sub" meant the _ character or the sub keyword...