Some researchers over at Johns Hopkins University have discovered that due to the way Variable Bitrate Compression does it’s thing, even if the audio stream is encrypted it is still possible to determine entire words and phrases based on the lengths of the packets with a high degree of accuracy.

According to the article referenced above it appears that the proof of concept tool is fairly limited, but given a little time and additional effort it’s capabilities could be greatly expanded, potentially to the point of transcribing entire conversations.

The researchers’ paper was presented at the 2008 IEEE Symposium on Security and Privacy a few weeks ago.