vocoder freq. spacing

[Martin Czech]

All manufacturers in the past seem to have designed the filter bank
with non-uniform spacing, ie. the frequencys are not evenly distributed
over the whole audio range but in a band from 200Hz to 8kHz. Lower or
higher frequencys are covered with a single lp/hp filter.  This may be
due to speech processing, where important formants are in this range.
But for multipurpose applications (not speech input) a more "musical"
evenly spacing, say a third (ie. a constant factor) seems to be more
suitable.  Example: If you feed a slowly sweeped sine vco (20-20000Hz)
into analysis and white noise into synthesis, this gives "noise
melodys". I think this will sound better in the case of evenly
distributed filters. I noticed this on the MAM-Vocoder demo in
"Keyboards" (Germany), last issue.  You can hear the gaps between
filters (obviously a high q design), as well as the said uneven
distribution. By the way: this vocoder is not expensive, quite
versatile, and seems to sound much better then Doepfer's.
(No, I don't get money from MAM).

Now, what will I loose, if I make the filters evenly distributed ?


Another idea: The high q design has a very sound of its own, if you use
it as a filter bank. I don't want to use filters with that high q but
may add an optional resonator with high q at the end of the filter chain, in
order to get the "typical" sound.

??


m.c.