Vocoder design

[Jeroen Proveniers]

> -----Original Message-----
> From: Martin Czech [mailto:czech at Micronas.Com]
> Sent: Friday, November 10, 2000 12:51 PM
> To: J.Proveniers at orga.nl
> Cc: synth-diy at node12b53.a2000.nl
> Subject: RE: Vocoder design
> 
> 
> 
> :::The Elektor vocoder features a voiced/unvoiced detector. 
> It consists of a
> :::lowpass filter (A) and a highpass filter (B). The output 
> of these are
> :::rectified/lowpassed and then some decision logic follows. 
> Something like
> :::when level of A<B then voiced; if A>B then unvoiced. If A 
> and B are both
> :::zero then last decision remains valid etc.
> :::
> :::I found the schematics of the elektor vocoder somewhere on 
> the web, don't
> :::know where.
> 
> I wonder why people built additional filters. There are allready
> a couple of bandpass filters with envelope detectors in the
> analyser, why don't take that information for voiced/unvoiced
> detection?

That's a good point. Probably because the elektor vocoder was a modular
design. One think I can think of is that you would need two more mixers;one
to sum the low band and one for the high band. In number of components this
would probably not differ much.
 
> btw.: from looking at spectra of different words, unvoiced/voiced
> detection seems to be pretty difficult. I could not find any 
> simple pattern.

The decision logic seemed quite complex to me when I took a look at the
schematic. In most cases it works well. Another technique it to mix some of
the high spectra of the original sound with the final output. I believe this
is patented from what I recall of the article.

JJ