[sdiy] Weird Vocoder project

[Magnus Danielson]

From: Ian Fritz <ijfritz at comcast.net>
Subject: Re: [sdiy] Weird Vocoder project
Date: Mon, 28 May 2007 10:00:03 -0600
Message-ID: <6.2.1.2.2.20070528095359.0220bef0 at mail.comcast.net>

> At 09:22 AM 5/28/2007, Magnus Danielson wrote:
> 
> >A typical vocoder analyzer channel will use a dual-resonance (4-pole) bandpass
> >filter where as a graphical EQ uses typically a single-resonance (2-pole)
> >bandpass filter. The dual-resonance peaks are spread such that you acheive an
> >approximation of a flat pass-band with a little responce-dip inbetween. The
> >slopes at the resonances are steeper than the bandpass roll-offs would allow,
> >and since the non-resonant gain is set to damping, those slopes is 
> >considerably
> >below the detection level. Those, response-wise they are quite different
> >animals.
> 
> Yes, this is a very important point.  If you use hi-Q but low-order filters 
> you may run into a problem because the low-order filter response has fairly 
> strong tails away from the resonant frequency.  When you have 30 of these 
> piled up you get a significant background signal.  I learned this from my 
> 36-stage fixed frequency filter bank. :-(

Yes, I was considering writing about this too. High Q automatically results in
longer release-time of the analyzer, and the transients is thus lost and then
the pronounciations will certainly be lacking the will to "follow along".

As always, to get better performace you need to spend more on filter
components.

Cheers,
Magnus