I was under the impression that running realtime voice-recognition on all the open phone lines at any one moment would require so much computing power that there would be no way you could hide it?
You'd need datacentre after datacentre after datacentre, surely?
Also, state of the art voice recognition is still flakey in a silent, echo free room using a condenser mic. Imagine how flakey its going to be going over chinese copper and GSM codecs. False positives ahoy! Even the chinese govt. has a maximum acceptable false positive rate for this kind of thing.
I haven't believed in this rumour all my life so far when it's been the US govt. claimed to have been doing it, and I don't believe it now that its switched to being the chinese govt. Not without some evidence.
It doesn't even make logical sense. Assuming you could actually do this, why cut the calls off? Why not record them as part of wider evidence gathering?
You could do fairly accurate real time voice recognition in 2000 on a home PC. Accuracy was and is an issue, but you only need to tag interesting conversations / people not perfectly transcribe everything. Now finding the lowest cost per FLOP is complex but to be safe let's say 1k of computing can decode 10 phone calls or 100$ / call. Now the amount of time people spend on the phone varies a lot, but let's call it 5% of the time on average to be safe. Edit: each call uses 2 people. Thus somewhere around 1 billion people * .05 / 2 * 100$ = 2.5 billion you could buy the hardware to transcribe everyone's phone calls in china. As to hiding things you could keep the computers anywhere so it would be fairly easy to hide.
PS: I would assume you could probably do it for a lot less but these numbers where just to show it's possible. Doing something like this in the US for ~750 million is going to be vary tempting, but I think the manpower costs for false positives may be the real issue.
My point wasn't that it isn't possible. I mean, who's gonna make that bet, especially on this website.
It was more that it would be impossible to do without it being flamboyantly obvious that you're doing it. I was referring to 'hiding' in terms of infosec as opposed to 'hiding' in terms of physically stashing away the boxes.
If they were doing it, we'd have heard from an engineer involved somewhere along the line, as opposed to rumours originating from paranoid citizens.
I'm sure the Chinese already have a network of informants and surveillance to gather "evidence". A system like this that just terminates calls is probably meant to slow the organization of mass protests. They probably figure that a near-spontaneous protest can be reduced in size or eliminated by making it take longer to get the word out. If people are angry about a specific incident, days or even hours of delay will make cooler heads prevail.
I doubt the Chinese would approach this in such a blunt manner. But I'm sure the Chinese government has lists of known and likely dissidents, and they probably have access to social network graphs for those individuals.
So you monitor likely dissidents who are hubs of major social networks more aggressively. For people who present randomly marginal political risk or low impact because of a limited social network, they get monitored less aggressively or with random sampling, etc.
China recently started requiring ID to get a mobile phone. They may be tracking only foreigners, and only those after they started to require ID. It actually may be that they have people listening in on calls, and it wouldn't take too much people (by Chinese govt standards) to go after a statistically significant portion of the population.
Anyway, if such a thing was implemented, I wonder if it could drive a set of language shifts. E.g., people stop saying "prison", they say "pickle jar"; they stop saying "informant", they say "cockroach"; etc.
GSM is actually already doing half the processing for you by reducing the dimensionality of the data. Pattern-matching a few dozen words using a custom neural network ASIC could probably be done in real time with very little hardware.
The calls are cut off because it's simple and effective. If people can communicate and organize, the damage is already done and you have to respond with force to stop protests. But if you can prevent protests from ever being organized in the first place...
>GSM is actually already doing half the processing for you by reducing the dimensionality of the data
Could you expand on this? Or just give me some key terms that I can look up to get more info. Sounds very interesting.
EDIT: Also, I don't buy the argument that dropping the calls is simple and effective. The antigovernment types are simply going to start using trivial codewords, like every criminalised person throughout history has done. Therefore it will barely touch them.
However it will drive businesses, both chinese and foreign, crazy. They are not interested in codewords. If you criminalise free speech, only criminals will speak freely.
Lossy compression involves throwing out redundant data; if you look at the data as a vector, you are effectively reducing the number of dimensions in the vector. Speech recognition involves rearranging the data to find its primary components (e.g. pitch, timbre, etc.). Most voice / audio compression schemes perform a similar rearrangement, as this is a convenient way to find (and eliminate) redundant data.
In particular, Fourier transforms, wavelet transforms, cepstrum transforms, subband filtering, and principal component analysis are techniques common to both audio compression and speech recognition.
Assuming they are mostly interested in preventing protests, I think the correct course of action would be to filter out most citizens.
Ignore the nursing homes.
Ignore calls that involve distance over 100 miles.
Ignore calls to contacts that theycall daily.
And escalate others.
Escalate Urban centers.
Escalate more than 2 phone calls per hour.
etc....
I live in China, so I did a quick test with my coworkers. 4 people. Two phone calls each. Each conversation we dropped the word 'protest' in both English and Chinese. No magical dropped calls for us.
Chinas telecom and internet monitoring and blocking is regional, what may be blocked in one region may be available in another, although there is certainly some centralisation of this system.
The goverenment is certainly experimenting with such a system in one or two regions, along with their pervasive cameras and face recognition.
However, seeing as the government here seems to have some difficulty building a road on time, on budget, that is actually decent quality it seems unlikely that they'll get this working.
Dear Chinese Government (and other petty dictatorships):
Your time is coming to an end. You can try to initimidate, and beat down dissent all you want. But in the end, freedom will prevail. But to hell with it, Neo said it better:
"I know you're out there. I can feel you now. I know that you're afraid. You're afraid of us. You're afraid of change. I don't know the future. I didn't come here to tell you how this is going to end. I came here to tell you how it's going to begin. I'm going to hang up this phone, and then I'm going to show these people what you don't want them to see. I'm going to show them a world without you. A world without rules and controls, without borders or boundaries. A world where anything is possible. Where we go from there is a choice I leave to you."
Who writes the software that enables phones to be tapped into? Even the U.S taps into phones, how is this feature protected? I imagine code has to be written to allow someone to listen in your convo, is that IN the droid OS or iOS?
The british arrested scholars for acquiring al qaeda literature. Cutting a phone call is quite child play in comparison but there's no sense of western ethical superiority mocking the british so it's less fun.
You'd need datacentre after datacentre after datacentre, surely?
Also, state of the art voice recognition is still flakey in a silent, echo free room using a condenser mic. Imagine how flakey its going to be going over chinese copper and GSM codecs. False positives ahoy! Even the chinese govt. has a maximum acceptable false positive rate for this kind of thing.
I haven't believed in this rumour all my life so far when it's been the US govt. claimed to have been doing it, and I don't believe it now that its switched to being the chinese govt. Not without some evidence.
It doesn't even make logical sense. Assuming you could actually do this, why cut the calls off? Why not record them as part of wider evidence gathering?