Amanda Walker wrote:

John Kelsey <kelsey@email.plnet.net> wrote:

This may or may not be relevant, but I was a math major during an NSA college- recruitment drive in the early 1980s. As part of the "come work for the NSA" packet they were giving out, they had a sample issue of the NSA internal linguistics journal. One of the articles concerned a system called DYPTRACK, which tracked the pitch of a digitized voice signal to an impressive degree of accuracy. This would imply that the NSA, at least, had invested considerable resources into digital speech analysis as early as the late 1960s or early 1970s (since by 1980 DYPTRACK was un-sensitive enough to be described in college recruitment material).

I have a CD today (used to be an LP in the late 1970's) of Enrico Caruso arias, digitally processed thru something called Stockham/ Soundstream digital process, circa mid-late 1970's. They took a number of samples of a modern (1970's) tenor's singing (someone whose voice had similar characteristics to Caruso's) and had the computer run it against the original Caruso recordings in an attempt to subtract out extraneous noise, resonances, and so on. The process helped somewhat, making the voice seem closer intead of somewhere a couple of rooms over, but the essential quality of the mechanically-recorded sound was still there, i.e., very dull and not at all life-like. They were supposed to release a lot more of this stuff, and they did some, but major interest was never there, and I haven't heard any more about further research in the intervening years, although I'll bet it's gone way beyond where it was.