Vocoder Idea

[Brian Patrick Towles]

Okay, I'll try to validate the idea of the "sweeping" vocoder a little bit
more.  First of all, the Nyquist theorem ( and common sense ) tells us
that we'll need to look at a frequency for half a cycle to correctly
detect its amplitude -- in practice this might need to be more like one
full cycle of the wave.  So, as stated by others, the sampling time at
each frequency is proportional to the inverse of the frequency.  So, for
low frequencies you could end up spending a lot of time waiting for the
half or full cycle.  Do common vocoder designs really go to 40 or even
100Hz? I haven't seen too many schematics.  Those frequencies are not
really present in human speech, but could be in other modulation signals.

Here're some possible solutions: fixed bandpasses for the frequencies that
take too much time and sweep the rest of the frequencies.  Mutltiple
sweeping units -- i.e. do the analysis in parallel.  Or borrow ideas from
wavelet transforms: low frequencies don't change as fast, so you don't
need to sample them as frequently.  That way could sample the higher
frequencies several times between each sampling of the lower frequencies.
I do realize that the control of this circuit might be harder than it's
worth.

Brian