5 ms·

> Instead of DTLS/SRTP, we decided to use the faster Salsa20 encryption. In addition, we avoid sending audio data during periods of silence — a frequent occurre

by GranPC 8y ago

> Instead of DTLS/SRTP, we decided to use the faster Salsa20 encryption. In addition, we avoid sending audio data during periods of silence — a frequent occurrence especially with larger groups. I wonder whether this would help an attacker infer voice data, with a method similar to the one from the paper "Uncovering Spoken Phrases in Encrypted Voice over IP Conversations" [1] [1]: http://www.cs.unc.edu/~fabian/papers/tissec2010.pdf http://www.cs.unc.edu/~fabian/papers/tissec2010.pdf

EvangelicalPig 8y ago
I've never tried doing pcaps with Discord voice chat yet but it would be interesting "estimating" what was said best on the amount of voice traffic transferred. Might be harder if it still looks like standard Discord TLS "control traffic" blending in though, no?
- wahern 8y ago
  Keystroke timing analysis has some solid research. See, e.g., Timing Analysis of Keystrokes and Timing Attacks on SSH (https://people.eecs.berkeley.edu/~daw/papers/ssh-use01.pdf https://people.eecs.berkeley.edu/~daw/papers/ssh-use01.pdf), Don’t Skype & Type! Acoustic Eavesdropping in Voice-Over-IP (https://arxiv.org/pdf/1609.09359.pdf https://arxiv.org/pdf/1609.09359.pdf). Deciphering content from latencies in packetized speech is likely much more difficult, but I wouldn't put much stock in it being too difficult. Which is to say, if you're transferring high-value information assets over VoIP you should probably assume it's decipherable. That doesn't mean you should change what you're doing. You could simple say, "M'eh, I'll worry about it when it becomes a thing." But I wouldn't assume it's confidential to someone willing to invest the time to target and capture the conversations. And I might leave a few choice comments in the source code and documentation so nobody could excuse imprudent reliance on confidentiality with, "But nobody warned me".
cvwright 8y ago
First author of the linked paper here. It's an interesting question. Actually the voice activation (on/off periods of sending vs silence) was the first thing we looked at in that project. There's definitely some information leakage there, but it was really hard to learn anything meaningful from it. The problem seems to be that long strings of words all get lumped into a single activation. It's really hard to discern anything about what those words are, or even what language they come from. We got some very weak results on language detection from VAD. You can learn something about the language, but it's not very precise. For example, maybe you could tell that a given conversation is definitely not language A, B, or C, but it might still be language X, Y, or Z.

by GranPC 8y ago

EvangelicalPig 8y ago

I've never tried doing pcaps with Discord voice chat yet but it would be interesting "estimating" what was said best on the amount of voice traffic transferred. Might be harder if it still looks like standard Discord TLS "control traffic" blending in though, no?

wahern 8y ago
Keystroke timing analysis has some solid research. See, e.g., Timing Analysis of Keystrokes and Timing Attacks on SSH (https://people.eecs.berkeley.edu/~daw/papers/ssh-use01.pdf https://people.eecs.berkeley.edu/~daw/papers/ssh-use01.pdf), Don’t Skype & Type! Acoustic Eavesdropping in Voice-Over-IP (https://arxiv.org/pdf/1609.09359.pdf https://arxiv.org/pdf/1609.09359.pdf). Deciphering content from latencies in packetized speech is likely much more difficult, but I wouldn't put much stock in it being too difficult. Which is to say, if you're transferring high-value information assets over VoIP you should probably assume it's decipherable. That doesn't mean you should change what you're doing. You could simple say, "M'eh, I'll worry about it when it becomes a thing." But I wouldn't assume it's confidential to someone willing to invest the time to target and capture the conversations. And I might leave a few choice comments in the source code and documentation so nobody could excuse imprudent reliance on confidentiality with, "But nobody warned me".

cvwright 8y ago

First author of the linked paper here. It's an interesting question. Actually the voice activation (on/off periods of sending vs silence) was the first thing we looked at in that project. There's definitely some information leakage there, but it was really hard to learn anything meaningful from it. The problem seems to be that long strings of words all get lumped into a single activation. It's really hard to discern anything about what those words are, or even what language they come from. We got some very weak results on language detection from VAD. You can learn something about the language, but it's not very precise. For example, maybe you could tell that a given conversation is definitely not language A, B, or C, but it might still be language X, Y, or Z.