Broadband Advisor

A common compression technique can make internet telephone calls significantly more susceptible to bugging, according to recent research from Johns Hopkins University.

Internet telephony has become widely used through consumer-centric applications such as Skype, and is becoming more common in enterprises.

The new research suggests, however, that standard encryption and compression methods, when used together, are not sufficiently secure.

VoIP calls are commonly encrypted using a technique that preserves the lengths of voice patterns in the original, unencrypted conversation, the researchers said.

Such calls are relatively difficult to listen in on, they said. But when length-preserving encryption is used with the variable bitrate (VBR) compression technique, the combination leaks a significant amount of information about the conversation, they found.

"Previous work has shown that combining VBR compression with length-preserving encryption leaks information about VoIP conversations," the researchers said in the report. "We show that this information leakage is far worse than originally thought."

In such conversations, particular phrases could be identified within encrypted VoIP calls with more than 90 percent accuracy. Even in decoding entire conversations, accuracy was significant, the research found.

"On average, our method achieves recall of 50 percent and precision of 51 percent for a wide variety of phonetically rich phrases spoken by a diverse collection of speakers," the researchers said.

The problem arises because VBR compression alters the compression rate based on the type of sound being compressed, using higher compression for simpler sounds and lower compression for more complex sounds.

The resulting audio stream makes spoken sounds easy to identify based on bit patterns, and these patterns are preserved in length-preserving encryption.

The researchers used a dictionary of spoken sounds to reconstruct the original conversation without the need to crack its encryption. The reconstructed conversation could then be analyzed as a full conversation or automatically scanned for particular keywords.

The technique could, for instance, be particularly effective in identifying certain predictable phrases used in professional business conversations, the researchers said.

The research analyzed the impact of noise, dictionary size and word variation on the performance of the technique.

"The results of our study show that an attacker can spot a variety of phrases in a number of realistic settings," the researchers said in the paper.

For mitigating such attacks, padding could be used to make the bit patterns less recognizable, the researchers argued.

However, none of the default encryption transforms of the Secure Real-time Transport Protocol, a standard for secure VoIP calls, specify the use of padding, the researchers pointed out.

"Although padding could introduce inefficiencies into real-time protocols, our analysis indicates that it offers significant confidentiality benefits for VoIP calls," added the researchers.

The paper (PDF), called "Spot me if you can: Uncovering spoken phrases in encrypted VoIP conversations", was presented by five Johns Hopkins researchers last month at the 2008 IEEE Symposium on Security and Privacy in Oakland, California.