Vocoder Design Questions, thanks for input

[Scott Bernardi]

Thanks, guys, for your input.

>think about thee phase relationships at the 'edges' of the bands.
>the question of what order filters to use for an efficient (male voice)
>vocoder was treated in the Journal of the Audio Society of America
>20 years or so ago.. sorry exact ref not to hand, but I remember
>the filters were NOT identical
>
>paul perry melbourne australia
>
1. A way to get around the phase reversals of the bandpass filters (which
would cause deep notches in the frequency response) is to alternate the
phase of the bands - inverting, non-inverting, etc. That way the point
where the response curves cross add instead of subtract.
2. How would you go about tracking down an old JAES article? I went to the
AES sight and did a search on "vocoder" and got three references, but I
couldn't tell if I had or hit or not.

>I've written some little c programms to compute
>component values for sallen&key and multi-feed-back filters (MFB), with
>given frequency and q.  These simple programms do also some variation
>analysis, ie. they change all component values a little bit around the
>optimum value. And this shows (as you can read in the books) that
>sallen&key filters with higher q are *very* sensitive to component
>value variations. This would mean to have at least 0.1% tolerance,
>which is very hard to do for capacitors. MFB filters are ot so
>sensitive but need on the other hand high opamp open loop gain for
>higher q values in the order of >>q^2. This is a problem at the high
>end of the audio band, the open loop gain of most op amps is not
>sufficient. NE5532 or OP275 would be a bit better.
>
>Martin Czech

I had planned on hand measuring the caps (my DMM has a 2.5% accuracy on
capacitance measurements, but it's the matching between caps that's more
important for the Q), then calculating the necessary resistance values
(yeah, I've written little computer programs also). If necessary, you can
even adjust resistance values by filing standard carbon resistors. So I
believe I can get around the tolerance issues.
I planned on a Q of 6, and the Sallen-Key gain is 2Q^2, or 72. My highest
bandpass was going to be at 3240hz, and figuring I need at least a factor
of 10 open loop gain above that, we have 720 at about 3200 hz. The TL074's
I planned to use have a GBW product of 3Mhz, with an open loop gain of 1000
at 3khz, so that should be good enough.
Anyway, I'll do some breadboarding to see what I can come up with
experimentally.


For anyone else interested in vocoders, check out the Elektor and ETI
vocoder articles (complete with schematics) on Anders Sponton's site at 
http://www.tellus.vallentuna.se/gymninfo/personal/anders/andersus.html.

Here's another tip I picked up from the Elektor article: One problem that
vocoders have is reproducing high frequency unvoiced (the hiss in "s",
etc), sounds when the carrier doesn't have enough high frequency content to
reproduce them. One solution is to include a noise source that is mixed in
the high frequency bands with the carrier when unvoiced content is
detected, to make up for the lacking spectral density in the high
frequencies in the carrier.
Another solution that was patented by Harald Bode in the 70's, is to just
mix the high frequency band of the excitation source to the output
amplifier. It's a lot simpler (a trimpot and a couple resistors) than
building voiced/unvoiced detectors, and is supposed to (according to the
article) increase intelligability immensely.

--------------------------------------
Scott Bernardi
voice: 415-538-0439 (note new number)
fax:   415-904-8375
scott.bernardi at fritz.com