Flabbergasted
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 01:19 AM
Original message |
NSA spying will never have a concrete role in counter-terrorism |
|
They are using computers, to check for keywords and do voice recognition. They will be looking for words like bomb and Allah in the same sentence supposedly. Which means an islamic extermist would have to be stupid enough to use these words on the phone. I've been told by several people over the years that they have a system already in place that picks up on words like pot or meth etc in order to find drug users. Which is why druggies rarely use these words on the telephone and only if they are ignorant. The obvious answer is to use symbolic words.
Voice recognition works only if they have a sample of someones voice. Which implies they would have direct information that this person is an islamic extremist. If they have time to get this information they have time to set a tap on this phone exclusively and get a warrant
It still comes down to spying on people who are constitutionally protected and should not be spied upon. Why do we let them frame this as a "terrorist" issue when its not? Its a control issue. Its a lets keep track of our political opponents issue. Its a lets keep track of protesters issue.
|
leftstreet
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 01:23 AM
Response to Original message |
1. BOMB ALLAH EXPLOSIVES ANTHRAX USAMA TERRORIST |
|
Come and get me, NSA.
I'm here all week.
:)
|
dusmcj
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 02:05 AM
Response to Original message |
2. you're spreading bad data: untrained voice reco is ubiquitous |
|
Edited on Tue Jan-24-06 02:07 AM by dusmcj
in automated telephone response systems. Call 800-326-3264 to find the Regal Cinema in your area. While the sequence of menus may drive you crazy, the voice recognition itself is pretty successful.
A random sample of credit card help lines and the like will reveal similar systems, all using untrained voice reco.
In short, voice recognition capability exists which can interpret the speech of the mass American public from an audio stream and turn it into text data, at which point it is searchable for keywords.
I'll guess that since voice reco is fairly intensive, harvesting a significant fraction of total domestic American phone call volume would require delayed processing, i.e. they couldn't put together enough compute power to do reco on the calls as they happen, but instead would have to copy the voice data for each call to storage to give themselves enough time to process it. Following that the search on the resulting text would be quick.
My current hobby thought experiment is to figure out how much bandwidth the American telephone system would require for its digitized audio data. I.e., for 300 million Americans making phone calls, how much storage would say 1 second of their phone calls require ?
Oh, by the way, while I work in IT I have no access to either classified data, or to particular information that the average Chinese, Russian, Israeli or Indian graduate student here on a temporary visa doesn't have. They designed and wrote all this stuff to begin with. I.e., for any sniffers reading this, don't trouble yourselves for a moment that I'm giving away state secrets here. Al-Qaeda seems to have folks doing its work who are at least as smart as I am, and probably went to graduate school here just like those nice folks I just listed. In addition, we can all read trade journals.
|
Flabbergasted
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 09:04 AM
Response to Reply #2 |
3. So in other words I'm full of it |
|
I thought voice recognition was referring to technology capable of recognizing a persons unique voice pattern?
|
dusmcj
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 01:13 PM
Response to Reply #3 |
4. you almost got it, keep at it |
|
it's technology that takes sound information (so basically the electrical signal that goes out to a speaker like in your phone) and looks up the sound patterns in a table which associates those patterns with the equivalent words (text data, like in a word processing file). For example, the signal pattern for the sound of 'cat' is associated with the text string "cat". Once you have the text string, you can search against it for words of interest, like "osama".
The point of our little disagreement here is that once upon a time (about 20 years ago) voice recognition systems needed to be trained as you describe so that an association table could be built for the unique sound patterns made by a particular speaker with the equivalent words. Since the pattern you make when you pronounce 'cat' is different than the one I make. The software technology (algorithms) were not sophisticated enough to generalize our two patterns so that it could recognize a word as "cat" when either of us, or in fact any member of the American public, said 'cat'.
That has changed. You can now encounter general voice recognition systems in many automated telephone response systems, for example credit card helplines and the movie chain listing system I mentioned. They're able to understand speech from any speaker within certain bounds, for example the credit card system can understand numbers, and the movie system can understand movie titles. Which points at how advanced the technology has become, since the movie titles change every week so that recognition of new words must be automatic and proceed according to phonetic rules as opposed to pure pattern matching. I.e. spoken 'syriana' produces a sound pattern which can be generally recognized from any American speaker to match to "Syriana".
I point at American public since I don't know whether the technology can deal with accents yet, or whether for example British English pronunciation would need a different rule set from American pronunciation. Similarly I would expect that foreign languages would need different rule sets. But maybe not, I don't know the state of the art.
At any rate, you weren't far off. (And I wonder whether you were thinking of voiceprint identification systems, where a speaker is identified by his unique sound formation patterns ? Similar idea, but there the aim is to identify the uniquenesses in his pronunciation, not average them out.) Keep up the interest and the activism, there are no wrong answers, only better ones.
|
Flabbergasted
(1000+ posts)
Send PM |
Profile |
Ignore
|
Tue Jan-24-06 05:59 PM
Response to Reply #4 |
5. Thanks alot for the info. I do like to learn. |
|
Yes I meant voiceprint ID, though I didn't know what it was called. I had not read that this was being used for NSA spying just the recognition.
|
DU
AdBot (1000+ posts) |
Thu Apr 25th 2024, 08:53 AM
Response to Original message |