Keep us posted... this is becoming a fascinating subject.
It's sounding more and more like John Woodgate is looking for a log amp. I
had a class on them once. ;-)
And to Mr. Dyson - thanks for your exposition; I "get" the concept of FFT
- turn the signal sideways, nothing to it! I just don't want to plow
through all that arithmetic. I once did a bank of 8 bandpass filters, much
like a "graphic equalizer", in an effort to extract formant information
based on the _relationship_ of the components. I saw an experiment on
edjamacayshunal TeeVee, where they did a Pavlov's Dog-style experiment,
with human infants. They used, instead of a bell, a human voice saying one
phoneme, for example, "ah", or "ee". They trained the human infant to
respond to a particular phoneme, and the infant responded to the same
phoneme, _no matter who said it_. Old man, young girl, child, adult man,
adult woman - it didn't matter, as long as they were saying the same
phoneme. So clearly, the phoneme itself is determined by the _pattern_ of
the various formants, without regard to the fundamental, or even the
specific frequencies.
I got bogged down while trying to find some kind of pattern-matching
algorithm, and had to abandon the project because of other considerations.
It was intended to be a speakwrite. Or a first pass at one. To control a
handicap-assist robot.
Thanks,
Rich
