Is there any reason why cubic splines are a poor method for
interpolation of audio signals such as speech or musical instrument
sounds?

Also, once upon a time I had a copy of a paper written by someone (if I
recall correctly) who worked for E-mu systems talking about pitch
shifting. Again IIRC, the conclusion was that for a given amount of
computing power, polynomial interpolation would always beat
decimation/interpolation as a pitch-shifting technique given the
measurement of distortion in the paper. I note that R B-J seems to know
most of the computer music papers I know, perhaps he recognises it.
What I would to revise is exactly how polynomials were fitted to the
samples in that paper. If no-one recognises the paper, then general
comments on the use of cubic splines would be appreciated.

Cheers,

Ross-c