AW: vocoder freq. spacing

[BJ]

Haible Juergen wrote:
> 
>         Equal vs. unequal spacing:
> 
>         First of all, an equal spacing would be the *easiest* way
>         to build a filter bank or vocoder, as it's easy to get
>         "equally spaced" components: You can use standard
>         E6 or E12 values. Calculation is then very easy. Of course
>         you end up at very few (Paia: 8) or very many (Synton: 20)

>         It's absolutely right that the MAM vocoder - as well as the
>         Roland vocoders - is specially tailored to speech
>         inteligibillity - so a number of 11, unequally distributed
>         bands are an optimum of excellent performance vs.
>         affordable price. Of course the building is not so easy
>         as with equally spaced bands, anymore: You can't simply
>         use E6 values and select them. The MAM solves this
>         problem the hard way, and the most excellent way: Each
>         filter has its own trimpot and is hand-tuned (!) by their
>         chief designer himself. (He won't let do this anybody else.)
> 
>         If we leave vocoder applications aside:
> 
>         Now, there's something general on equally spaced
>         filter banks: You have the choice to set the bands to
>         musical intervalls that are fractions of an octave, or not.
>         If you tune them in octaves, of fractions of an octave,
>         and feed it with a sweeping oscillator, you will get
>         a very uneven response: Every time your VCO's
>         fundamental hits the resonance frequency of one band,
>         many of the VCO's harmonics will hit exactly the resonance
>         frequency of another band. So you have very strong
>         resonance in one moment, and almost nothing in the next
>         moment of your sweep. This behaviour may be considered
>         pleasant or unpleasant, depending on your application.
>         Different manufacturers took different different paths here.
>         The EMS filter bank is tuned in Octaves, for example.
>         The coresponding Serge Module (don't know the exact
>         name) has its bands deliberately *not* set to an octave
>         (was it a 7th ? - don't remember). So there would be
>         a greater variety of hormonics hitting individual resonance
>         frequencies.
> 
>         So, as usual, everything has its uses.
> 
>         JH.

Good and informative posting here juergen.

My point to this vocoder thing is dont callculate your self to death.
Since its easy to do it with vocoders.....
One solution tho this resonance thing ist to input as much signal as
possible
into the vocoder and use a compressor/limiter to help you out.
Almost every analogue vocoder uses a compander some vocoder uses
multiple companders (Roland SVC350 for example).
My solution to the "equally spaced BPF bands" as I use in my VOC24
vocoder
is to use a standard series of capacitors and then use some variations
in the use of resistors from different series.
This is a wery usable way to do it since its farmore easyier to get hold
of different standard series resistors then to use strange values on the
capacitors.

Designing ONE vocoder for your self is one matter,and designing a
vocoder for
sale in small series are a whole other matter.

Please read what Dieter Doepfer has to say about vocoders at my
"vocoder info page" at: http://www.algonet.se/~zzynt
and have a good moment of thinking about your d-i-y vocoder design.
You will also find the Voc24 vocoder there as well.

And speech inteligibillity are a big issue among the analoge vocoders,
a vocoder whitout a Unvoiced/voiced detector/processor are very
dull sounding and you need to have very very narrow BPF's and plenty
of BPF channels to get any usable sound out of the unit at all.
So therfore use a  UN/V detector/processor for your d-i-y vokkoder..
  
BJ