[sdiy] Anyone built the ELEKTOR VOCODER?

[Dave Krooshof]

>Here is a idea for a different approach.
>Have a 30 band spectrum analyzer on the  modulator input.
>And use only 3 tunable filters on the reconstruction/carrier side.
>
>When the reconstruction filters are tuned in real-time to the vowels that
>the spectrum analyzer detects,
>it should be possible to get good understandable speech form only 3 filters.

This is actually the most beautiful way of building a vocoder, as tuning
filters is what happens in a mouth. Fixed filters that open up, is just
voice-ish.
Three moving filters is the real thing.
So the trick is to find the highest peaks in the spectrum, and tune a
filter accordingly. A slew of the response defines the vocal quality.
The downside of it is that you need a rule to judge by when you are about
to close a filter. Otherwise, they just jitter about in non voice/low
volume input from your mike.
I found that you did not need to set volumes in each filter, Q factor is
more important. For volume, you can live with one overall volume control.

A FFT in the digital domain might give the quickest results.
If I were to build it again, I'd use MaxMSP.

In the analogue days (BEA5! @ http://www.koncon.nl)
I did it by filtering the controlsound into three bands, slightly overlapping.
The output of each of these bands were fed into a Frequency to voltage
convertor.
The outputs from this machines were fed into three filterfreq inputs.
This gave very nice results, after setting the Q's.
Very vocal, and understandable in english (not in dutch due to the lack of
gggrkhkhk sounds from the patch!)

After this I came to the conclusion that vocoders are dull in themselves.
I have nothing to say through a vocoder.
So I made a patch that would make the studio brabble by itself when faders
were moved (tone on a channel, Volume to voltage, voltage to network that
wanders about before allowing a new stability=silence)

A studio talking in it's own language is a lot nicer to be in.

And oh yeah, now I remember:
I took a sine clipped on one side for the sound to mudulate.
For utter voicyness, I skipped a period here and there, randomly, this
happens in voices too, say at a 1 out of 15 or 20 rate.
For feminity I added a little pink(:-) noise.

>How about hacking an already existing stereo
>equalizer?  You just need to disconnect
>the filter bands from the common buss,
>add diode and cap DC converter to input filter bands,
>and insert VCAs on the output filters.  Add a sibilance filter and noise
>source if desired.
Mwoh:
Too broadbanded filters. Too much overlapping!


dave

--------------------------------------------
Dave Krooshof http://www.xs4all.nl/~krooshof
geluidstechnicus @ http://www.ahk.nl/the/theatertechniek_ov.html
webmaster: http://www.popronde.nl
-------------------------------------------