I just came accross an interesting attack on the Secure Real Time Protocol (SRTP) using Variable Bit Rate codecs (VBR). That protocol is used to secure voice or IP communication by encrypting the transmitted data.
It uses the phoneticpronunciation of words to identify patterns in the VBR encoding which can be used to bypass encryption and identify phrases as well as the language spoken. So this attack does not target individual words, but phrases or sentences.
Although the paper claims in some cases a success rate of 90% it has an average of 50% success, which is already good enough! What is the answer to this issue? padding!