newton2@ucbtopaz.CC.Berkeley.ARPA (01/16/85)
This is a renewed request for discussion/comment/pointers on the subject of speech scramblers which work by temporily shuffling blocks of speech. If we assume the underlying cryptography to be secure (i.e. the generation of scramble-sequences throughout the message cannot be anticpated wihout the key), what are the techniques for cracking the system by direct assault on the analog scrambled signal? How quickly can this be done- Real time? I assume a digital random access memory- no clues from the differing noise level along the length of an analog shift register. However, there might be clues from the excitation of the (known) impulse response of the band-limited voice channel. And of course the cleartext (voix humaine) has some known characteristics. Is such a scrambler of *any* value against a determined opponent-- i.e., might it provide at least some time-limited security or require such special equipment for routine cracking that the use of large numbers of such scramblers might overwhelm a particular opponent (not knowing which messages were high-value?) I'd greatly appreciate any discussion of this, public or private. Thanks- Doug Maisel 415 549-1403
ron@brl-tgr.ARPA (Ron Natalie <ron>) (01/18/85)
The answer is not very. Some language and hearing researchers did some research into this and you have to do a lot of scrambling over a large time period to make it stop sounding like speech. Instead of shifting in the time domain as you are suggesting, why not (since your attack is to convert to digital anyway) encrypt the values? -Ron I'll try to find my reference material on this subject.