It took me a while to realize that gmaxwell was talking about secretly recording the audio interference from a nearby compromized computer which would then be retrieved and decoded by the attacker at a later time.
Actually, from the computer the Trezor is connected to itself (after all, the point of the Tezor is the assumption the host computer is compromised); nearby would be an even more impressive stunt-- in terms of nearby but not connected; that would better be done with a software defined radio receiver (e.g. not merely a remote compromise).
All of that is a long shot, but thats the annoyance about defense; you have to defend against all attackers, and an attacker may spend a lot of time and resources on a single valuable target. It's quite hard to be confident that you do not have an exploitable weakness. If you're sure you're secure you're probably not being creative enough.
wooow. crazy shit. reminds me of the hollywood movie "eagle eye". But not long ago i ve read an article about some scientists from Tel Aviv University who were able to extract RSA keys from the "noise" of a CPU!!! So the above scenario, where a nearby computer "attacks" a victim within his noise-recognition area by "listening to its processor", should be practically possible. maybe we should start adding noise protection measures to our computers and trezors... LOL