rbw wrote:
O.K. I'll bite...
Help me imagine what that calculation would look like...
;^)
rbw
Randall Shimizu wrote:
Well just imagine how much processing power it require to
simultaneously interpret 100,000 phone calls at once.
Oh, come on. I presumed that everybody on this list can do estimation.
Okay, human voice has most of its information content below 4KHz.
Sampling that requires at least 8Khz. Call it 10K samples/second.
We'll work with 1 bytes per sample even though 2 bytes is probably better.
So, 10^4 bytes per second per phone call. With 10^5 phone calls, that's
10^9 bytes per second *continuously*-1GB in traffic continuously.
Now, we have to analyze that data. An FFT can be done in O(n log n).
So we need 10^9 * log 10^9 flops. Or, roughly 10^10 flops continuously
producing frequency bins.
Now, we have to analyze the DFT's and convert them to something useful.
Hidden Markov Models seem to be on the order of O(n^2), so we go from
10^10 to 10^20. You need 10^10 (10 billion) computers operating at
10^10 flops (10 GHz) to chew through all of the data.
A little outside of even Google's ability to handle in real time. And I
haven't even mentioned power.
Now, we may not be producing 10^10 frequency bins, but the markov models
are not always n^2, in general. They are normally O(n^t) where t is
related to the number of identifiable phonemes. If t is 3, then 1
million frequency bins are active to get the same level. If t is 4,
then only 10000 frequency bins, etc.
In short, it's big.
You'd probably be better off hiring 100,000 people to listen in.
Assuming $10/hr x 40hr/wk *50wk/year thats only $20,000 per person per
year or $2 billion total.
This is why I'm so annoyed about the FBI wiretapping program. They're
not analyzing the data in advance except for a *very* small number of
people (and they're probably doing that with people). All they're doing
is recording it for use in fishing expeditions afterward.
-a
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
