[PD] IPPASS finds missing fundamentals

[katja]

Hello,

This is for math freaks. Last week I invented a function to help
detect period length in the case of missing fundamental. Well,
invented? I am not a professional math person, and it may be in use
since long, without me having heard about it. Anyhow I tagged it
IPPASS, Instantaneous Power of Phase-Aligned Signal Segment. The
function describes the squared amplitude modulation of an analytic
signal segment in which phases are aligned via the amplitude spectrum.
It happens to be the case that these modulations have single
periodical peaks exactly in the cases where the autocorrelation
function gets messed up by a dominant second harmonic. IPPASS could be
useful in time domain pitch processing, I'm planning to try that
shortly.

IPPASS is illustrated with plots and Pd demo patch here:
http://www.katjaas.nl/ippass/ippass.html

A couple of years ago, when figuring out an attack detector, I was so
much plagued by periodic amplitude modulations that they were still
fresh in mind when I delved into the matter of period detection
recently. That's how I got the idea to exploit the modulations this
time.

Does anyone know if this function or something similar is in use for
pitch tracking or another field of signal analysis? And another
question: does anyone have ideas to improve stability and accuracy of
a signal segment with it's phases forced to zero? It suffers from the
usual boundary effects, which must be reduced to a practical minimum.


Katja