[sdiy] pitch to voltage, pitch shifters

[Dave Krooshof]

>The stiffness of the strings makes the overtone
>series sharp... so adding the "right" overtones
>might be impossible.
For pitchtracking, that was not too much of a problem.
It was just a little flat.
>One problem of the added harmonic method
>is that some very real sources (my guitar
>for instance) have _non-harmonic_ overtones.
These are a problem. When I was in Sonology, we programmed
several pitchtrackers and FFT fun. That is, we came up with
ideas, and our teacher programmed in Delphi while we were
discussing the implementation.
An FFT can look messier then you might assume. In the digital
world you could throw the non correlating overtones out.
Conventional methods throw overtones out of the FFT when
they are below a treshold. But since our hearing doesn't do that,
but something way more complicated: Pattern recognition, throwing
non-correlating overtones out should give better results.


>I think the result would lead to intolerable IM disortion.
Sorry, what is IM? interference modulation or something?

>Sort of the same problem as using a multiplier to
>get a 2X frequency shifter... it only works on a
>pure sine wave.
Yes, because that's actually peak counting, or counting zero crossings.

>But hey I hope I'm wrong. I want a perfect shifter as well....
Would you be happy with a laptop on stage?

Is there a Max or -even better for this stuff- Super Collider
wizzard on this list? I think the super collider enviroment would
be an excellent place to go, unfortunately, I never managed
to get a finger behind it myself.
SC is capable of granular synthesis, so that's perfect for us.
(time frames and frequency content processed independant)


Let's define what a perfect pitchshifter would look like:
- It works instantly
- It does not jump octaves
- It can be DIYed in a rugged box
- It doesn't sound like a cheapo digital effect
- It can do pythagorean deviding and multiplication
- It works with low pitched sounds too.
It would be really perfect if several shifts can be done
in a row without getting too weird.


Can't find my filter cook book. I know it discusses "all pass" filters too,
but... Harry, would a All pass filter as discussed in an earlier mail be
possible?
That is, with equivalently less delay on higher freqs?

Are there chips that do FFT "just in time"?
Maybe we ought to hack a Minidisk walkman chip. It is said to contain
a pseudo neural network doing FFT. Then that data is stored on disk.
Fourier synthesis is done (alt least in walkmans) by driving some 64
oscillators by telling them digitally which freq, phase and amplitude they
should have. (This was an explantion I got when asking a MD techy why
MD walkmans never have a dig out)

So, maybe if we can obtain those two chips we could do this:
guitar -> dac-> FFT -> hack the data and do multiplication on each
frequency -> FS.
Pitch changing will not change the time, as the frames remain constant.
The grane/frame length might show though...
Would it be quick enough?

I'll open my MD and google for the chips....

Dave


Actually, I can NOT understand why none of the mp3 software and
non of the MD players is capable of pitchshifting, or even simpler:
time compression/expansion without changing the pitch.
Dansers would be so happy with that.