Now this is a thought provoking blog entry.
Freedom to Tinker found a paper from the University of California, Berkeley that posits that listening to someone typing on their keyboard, you can reconstruct what they type based on the different sounds the keys make, presuming they are typing english.
It seems very plausible to me, because I've noticed how keys sound different before. (Large keys versus standard size keys; frequently used keys versus less frequently used keys; outside keys versus inside keys; keys hit with strong (first/middle) finger versus keys hit with weak fingers.) Putting this together with known facts about english language such as letter frequency, common letter combinations, this article could have a potential impact on password security in closed environments.