"Phase-Preserving Narrowed Autocorrelation"?

[Fabien Gouyon <fgouyon@xxxxxxxxxx>]

List,

In "Computational auditory pathways to music understanding" (1997), Barry
Vercoe proposes the use of the Phase-Preserving Narrowed Autocorrelation
(PPNAC) for detecting recurring patterns and estimating tempo in a
representation of the loudness of audio data (1 value each 10 ms it
seems).
However there's no indication on what that is exactly, how it is computed
and where the "phase-preserving" feature resides in. Another feature of
the PPNAC seems to be to let large time scale phenomena influence smaller
time scales (e.g. a quarter-note pulse would induce a weaker response at
eighth-note, triplet-eighths, sixteenths, etc.). I didn't find details on
that either.
Would somebody have an idea of where to look for more detailed
information? Or maybe a hint to understand? Would there be some CSound
code around?
Thanks
____________________________________________

Fabien Gouyon
Music Technology Group IUA-UPF Barcelona
http://www.iua.upf.es/~fgouyon
tel : (00 34) 93 542 28 64
____________________________________________