modules with different tubes?(oktavers)

[Dave Krooshof]

>> 4. Components analysis. I got the idea from sociology, been programming
>> on it with my DSP teacher in sonology. Use it, but give me credits.
>> The idea is: Go in to the spectral domain, FFT.
>> Then consider the peaks.
>> Roll up the frequency axis, one oktave per revolution.
>> Now, when periodicy happens, that is rich in harmonics, it's peaks will
>>add up
>> on the oktaves and on the fifths, mostly!

Harry the man:
>This is similar to the method used by Axon in their "neural net" pitch
>detector. They take a number of delayed samples and look for correlation
>of the harmonic spectra. Basically they are looking for the echo pattern of
>the pick attack reflecting down the string. They claim less than 10mS
>response.
>Thats very good.

It isn't the same, it's more of a shortcut.
Come to think of it, it might be posible using short delays only instead of
FFT's.
Auto correlation might do the trick. It will take some math to do it though...
Kasper Sauren, do you have any hints?

Pitchtracking will always take some time, as a pitch is something that
happens in time. The real goal is tracking pitches before the even managed
to do a complete period.
Actually you can prove this both possible and unpossible:
After one period, you stil do not know wheather that was just a noise with
a wavy coincidence, or a part of a tone.
But listening to a very short click (say 1/3 period) give the ear an idea
of the harmonic content of that click. Fourier would have said this is
happening because describing the wave shape of this peak will need the
longer periods to add up. The downside however, is that the windowsize will
be so obvious in the caclutions.

Kasper, If your interested in these fenomena, you'll have some pretty
interesting mathtalks with Peter Pabon. next year.

--------------------------------------------
Dave Krooshof http://www.xs4all.nl/~krooshof
geluidstechnicus @ http://www.ahk.nl/the/theatertechniek_ov.html
webmaster: http://www.popronde.nl
-------------------------------------------
no html mail please