http://www.cs.unc.edu/~amw/resources/hooktonfoniks.pdf Phonotactic Reconstruction of Encrypted VoIP Conversations: Hookt on fon-iks Andrew M. Whiteb Austin R. Matthewsbb Kevin Z. Snowb Fabian Monroseb bDepartment of Computer Science b Department of Linguistics University of North Carolina at Chapel Hill Chapel Hill, North Carolina { amw, kzsnow, fabian } @cs.unc.edu, armatthe@email.unc.edu Abstract In this work, we unveil new privacy threats against Voice-over-IP (VoIP) communications. Although prior work has shown that the interaction of variable bit-rate codecs and length-preserving stream ciphers leaks information, we show that the threat is more serious than previously thought. In par- ticular, we derive approximate transcripts of encrypted VoIP conversations by segmenting an observed packet stream into subsequences representing individual phonemes and classifying those subsequences by the phonemes they encode. Drawing on insights from the computational linguistics and speech recog- nition communities, we apply novel techniques for unmasking parts of the conversation. We believe our ability to do so underscores the importance of designing secure (yet efo,cient) ways to protect the cono,dentiality of VoIP conversations. ... VII. CONCLUSION In this paper, we explore the ability of an adversary to reconstruct parts of encrypted VoIP conversations. Specif- ically, we propose an approach for outputting a hypoth- esized transcript of a conversation, based on segmenting the sequence of observed packets sizes into subsequences corresponding to the likely phonemes they encode. These phoneme sequences are then mapped to candidate words, after which we incorporate word and part-of-speech based language models to choose the best candidates using contex- tual information from the hypothesized sentence as a whole. Our results show that the quality of the recovered transcripts is far better in many cases than one would expect. While the generalized performance is not as strong as we would have liked, we believe the results still raise cause for concern: in particular, one would hope that such recovery would not be at all possible since VoIP audio is encrypted precisely to prevent such breaches of privacy. It is our belief that with advances in computational linguistics, reconstructions of the type presented here will only improve. Our hope is that this work stimulates discussion within the broader community on ways to design more secure, yet efo,cient, techniques for preserving the cono,dentiality of VoIP conversations.