AF developing DEA Wiretap Echelon-like Development Projects
Bill Stewart
bill.stewart at pobox.com
Tue Oct 8 18:09:49 PDT 2002
The following web page is about recent projects at the
Air Force Research Laboratory. Item 8 is about new wiretap technology,
designed to monitor large numbers of conversations for drug activity.
The accompanying artwork has a large and small version of a
wiretapper logo, which should be possible to abuse for something :-)
http://www.afrl.af.mil/accomprpt/may02/accompmay02.htm
Google cache:
http://216.239.53.100/search?q=cache:wHZ4vsieDNkC:www.afrl.af.mil/accomprpt/may02/accompmay02.htm+nanosat+2002&hl=en&lr=lang_da|lang_nl|lang_en|lang_fr|lang_de|lang_is|lang_es&ie=UTF-8
Wiretapper Logo: http://www.afrl.af.mil/accomprpt/may02/images/may_8.gif
The Information Directorate's Multisensor Exploitation Branch and Research
Associates for Defense Conversion (RADC) jointly developed, tested, and
demonstrated an experimental model capability that automatically extracts
information from telephone background sounds and conversational speech to
identify drug networks and the participants. The work, sponsored by the
Drug Enforcement Agency (DEA), addresses the problem of monitoring large
numbers of telephone conversations for drug activity, while protecting the
privacy of citizens in accordance with wiretap laws.
The capability called Automated Title Three Audio Correlation (ATTAC) makes
it possible to automatically segment and flag drug- related activity and
identify its participants without understanding the message content of the
conversation. A background sound recognizer technology identifies sounds,
such as dial tone, number dialed, ringing, and other sounds, while a Vector
Quantization speaker-recognition technology identifies the persons involved
in the conversations.
The DEA and RADC collected a database of conversations through 74
individuals who made over 1300 calls from cellular phones, and office
phones, and who used message machines. The results in identifying the
participants in conversational speech varied widely. DEA and RADC
technicians obtained good results (90%) when individuals used the same
phones; however, when the same individuals used different phones, the
performance could drop to as low as 55%.
The directorate is conducting research work to improve recognition across
multiple phone types. Although the directorate developed ATTAC for DEA use,
the technology developed advances the state-of-the-art in speaker
segmentation and in information extraction for the Air Force intelligence,
surveillance, and reconnaissance mission. (Mr. S. E. Smith, AFRL/IFEC,
(315) 330-7894)
More information about the cypherpunks-legacy
mailing list