Re: [AH] Synthesizing vocal sounds

From ndkent
Sent Sun, Jan 30th 2000, 03:19

> Well, a vocoder consists of an analyzer and a synthesizer (filter)
> stage. IMHO cheating would mean using the analyzer stage. Without that,
> you would have to build phoneme tables as EGs, and feed these into the
> vocoder filterbank instead of the analyzer CV. Early commercial computer
> speech applications used that approach heavily. Of course, doing it all
> analog would be very gear heavy - you'd at the very least need some
> 50-60 banks of n (10-20 for common vocoders) multistage EGs or sequencer
> banks each to get a minimal phoneme set that would be able to utter some
> comprehensible sentences. Some kind of digital storage of the envelopes
> certainly is warranted there.   

> True.
> But the nice thing about VCFs is that they can have more variable
> parameters, making them "parametric" filters.
> That way, you may only need 2-3 VCFs & a few VCAs for gain & resonance
> The di-phones (transition between 2 sounds) works better that way, too.
> I've used both the 'graphic eq" (Fenix) and the "parametric" (EMS
> filterbank) approach. Graphic is easier to tweak, parametric sounds better
> in my setup.

yes, there is no real need to use a pile of fixed filters, you really
need band pass filters

and can't you take the Yamaha FS1R approach into the analog domain. They
use just a handful of voiced and unvoiced formants being sequenced

here are some formant frequencies to vowels in the middle of this
doepfer page

nick kent