7 ms·

Voice-matching technology was developed by MIT/Lincoln Labs under NSA contract [pdf]

60 points by staunch 9y ago

Zigurd 9y ago
I'd bet this isn't the first implementation of speaker identification in surveillance networks, by a couple decades. It has always been easier to distinguish between speakers than is has been to transcribe speech, or even recognize isolated words, without training the recognizer to a speaker's characteristics.
drharby 9y ago
Similar thing happened with lens development at the nro and nasa
musorschet 9y ago
It would be nice if people would refrain from posting direct links to classified documents, even if they are on the Internet for all to see. For folks who actively hold a clearance and would prefer to continue doing so, having a classified document get cached on their machine could be an annoyance they don't want to deal with. It would be nice if it was clear from the post title that the document might be one to avoid downloading (or, should be downloaded with care).
rdtsc 9y ago
Once you can encode the voice characteristics to match it, then go to the next level and generate the voice so that it sounds like that person. That might be interesting in breaking ZRTP protocol which relies on verbally comparing codes both sides see on their ends. https://en.wikipedia.org/wiki/ZRTP https://en.wikipedia.org/wiki/ZRTP So when Alice calls Bob and they verify the code, she sees "123" and says "123". Eve captures "123" then used Alice's generated voice and tells Bob the code is "456". Then does the same for Bob's side. Now, Alice and Bob if they are suspicious they can try to repeat the code a few times, maybe grouping digits differently.
- lallysingh 9y ago
  Do it over video.
  - prutschman 9y ago
    https://www.youtube.com/watch?v=BoncWqttN6U https://www.youtube.com/watch?v=BoncWqttN6U I can only imagine it's gotten better in the past 20 years.
    - lallysingh 9y ago
      This is probably the state of the art: https://www.youtube.com/watch?v=9Yq67CjDqvw https://www.youtube.com/watch?v=9Yq67CjDqvw But still easy to tell if it's someone you know, and try to have a regular conversation while doing the video chat.
- rahimnathwani 9y ago
  According to Wikipedia "In late 2006 the US NSA developed an experimental voice analysis and synthesis system to defeat this protection"
  - rdtsc 9y ago
    Yap they tried but it was experimental. They want something practical. Well they might have solved the problem but I kind of doubt it. Notice that instead of a number it could be a set of random dictionary words. So instead of 123 they get "potato airplane". Now Alice and Bob start making jokes or comments about how silly that sounds and how far potato airplanes could fly etc. Now the problem of simply transcribing individual numbers got many times more complicated.
rurban 9y ago
I heard the british counterpart (GCHQ) had better voice matching in those days.
etiam 9y ago
Good to see this is getting attention. I was hoping this release would as well: https://theintercept.com/2018/01/19/voice-recognition-technology-nsa/ https://theintercept.com/2018/01/19/voice-recognition-techno... https://news.ycombinator.com/item?id=16193246 https://news.ycombinator.com/item?id=16193246
Dowwie 9y ago
If this is accurate, then so too is "my voice is my passport" type of voice printing authentication. I need to watch sneakers again..

60 points by staunch 9y ago

I'd bet this isn't the first implementation of speaker identification in surveillance networks, by a couple decades. It has always been easier to distinguish between speakers than is has been to transcribe speech, or even recognize isolated words, without training the recognizer to a speaker's characteristics.

drharby 9y ago

Similar thing happened with lens development at the nro and nasa

musorschet 9y ago

It would be nice if people would refrain from posting direct links to classified documents, even if they are on the Internet for all to see. For folks who actively hold a clearance and would prefer to continue doing so, having a classified document get cached on their machine could be an annoyance they don't want to deal with. It would be nice if it was clear from the post title that the document might be one to avoid downloading (or, should be downloaded with care).

rdtsc 9y ago

Once you can encode the voice characteristics to match it, then go to the next level and generate the voice so that it sounds like that person. That might be interesting in breaking ZRTP protocol which relies on verbally comparing codes both sides see on their ends. https://en.wikipedia.org/wiki/ZRTP https://en.wikipedia.org/wiki/ZRTP So when Alice calls Bob and they verify the code, she sees "123" and says "123". Eve captures "123" then used Alice's generated voice and tells Bob the code is "456". Then does the same for Bob's side. Now, Alice and Bob if they are suspicious they can try to repeat the code a few times, maybe grouping digits differently.

lallysingh 9y ago
Do it over video.
- prutschman 9y ago
  https://www.youtube.com/watch?v=BoncWqttN6U https://www.youtube.com/watch?v=BoncWqttN6U I can only imagine it's gotten better in the past 20 years.
  - lallysingh 9y ago
    This is probably the state of the art: https://www.youtube.com/watch?v=9Yq67CjDqvw https://www.youtube.com/watch?v=9Yq67CjDqvw But still easy to tell if it's someone you know, and try to have a regular conversation while doing the video chat.
rahimnathwani 9y ago
According to Wikipedia "In late 2006 the US NSA developed an experimental voice analysis and synthesis system to defeat this protection"
- rdtsc 9y ago
  Yap they tried but it was experimental. They want something practical. Well they might have solved the problem but I kind of doubt it. Notice that instead of a number it could be a set of random dictionary words. So instead of 123 they get "potato airplane". Now Alice and Bob start making jokes or comments about how silly that sounds and how far potato airplanes could fly etc. Now the problem of simply transcribing individual numbers got many times more complicated.

rurban 9y ago

I heard the british counterpart (GCHQ) had better voice matching in those days.

etiam 9y ago

Good to see this is getting attention. I was hoping this release would as well: https://theintercept.com/2018/01/19/voice-recognition-technology-nsa/ https://theintercept.com/2018/01/19/voice-recognition-techno... https://news.ycombinator.com/item?id=16193246 https://news.ycombinator.com/item?id=16193246

Dowwie 9y ago

If this is accurate, then so too is "my voice is my passport" type of voice printing authentication. I need to watch sneakers again..