Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!yale!think.com!samsung!uunet!motcid!duerr
From: duerr@motcid.UUCP (Michael L. Duerr)
Newsgroups: comp.dsp
Subject: Re: US Gov. sends real-time audio at 2400 baud!
Message-ID: <4481@apricot30.UUCP>
Date: 12 Nov 90 16:31:21 GMT
References: <1990Nov8.210640.2893@ux1.cso.uiuc.edu>
Organization: Motorola Inc., Cellular Infrastructure Div., Arlington Heights, IL
Lines: 25

From article <1990Nov8.210640.2893@ux1.cso.uiuc.edu>, by sl35746@uxa.cso.uiuc.edu (By-Tor):
> Hi.  I heard that the US government uses 2400 baud modems to send real-time
> audio, to get a secure line anywhere in the world.  Does anyone know what
> kind of algorithm they are using for compression? 
> Anyways, consider that 2400 baud=2400 bits per second, meaning
> without compression, you could get 2400 hz sampling at 1 bit resoultion.
> This is totally unintelligable, so how the heck could they compress it
> so much?  

From my experience, I would guess that they are using linear predictive coding 
of some sort. I also know that years ago the US Gov't had some systems that
used subband coding.  However, LPC seems to be the scheme most preferred,
probably for cryptological reasons that make the encryption hard to break.
LPC works by modeling the vocal tract.  No attempt is made to compress 
samples: this method works only on speech. 
Also note that baud defined as the number of transitions per second.
If each possible symbol has 4 or 8 possible values, for instance, the
bit rate would be 4800 or 9600.  Modulations like QPSK, QAM, MFSK, etc.
have these type of symbols.
As a last note, I would guess that whatever comes out of the spook 
boxes is intelligible, but not necessarily what would be considered
quality speech.
A lot of good work on speech compression appears in back issues of the
Bell System Technical Journal; also, I would recommed the book by
Oppenheim and Lim.