Vocoder Idea

[Magnus Danielson]

>>>>> "MC" == Martin Czech <martin.czech at intermetall.de> writes:

 MC> By the way: Does anybody know of other approaches to synthesis than a
 MC> multi bandpass structure ? I mean, it sounds really stupid to have 20
 MC> high q bandpass filters in order to emulate the vocal tract, which
 MC> seems to have only 3 or 4 formants. Linear predictive methods try to
 MC> find out the formants of the incoming signal and to adapt a few
 MC> variable filters (a,q, and f).
 MC> Ie. a clever analysis section may save synthesis overhead.
 MC> But I doubt if an analogue sulution would be feasible.

The type of formants that the human voice uses could be synthesised
with CV-controlable EQ-type filters. They are really that simple.
It is quite feasable to implment the synthesis part in analog. You
would save components as compared to the multichannel synthesis
section of a vocoder.

The analysis section is a diffrent story, and this is also why the
synthesis part locks like it does, since it needs to match the
function of the analyser chain. It would in a way be possible to
extract signals similar to the once needed for the EQ-approach.

There is a lot of literature in both voice synthesis and
analysis/recognintion. I have a book on "Theory and Application of
Digital Signal Processing" which in a chapter covers the
implementation issues on digital vocoders, all complete with
description on how the human voice is built up and all that. Compelte
with X-rays and anatomic drawings.

Cheers,
Magnus