An Efficient Method for Pitch Shifting Digitally Sampled Sounds

Abstract

Altering the pitch of digitally sampled sounds is an operation that is performed by a number of commercially available devices (sampling synthesizers, harmonizers, vocoders) as well as by the software of many computer music languages. However, the algorithms for resampling and formant analysis/ resynthesis that these devices implement have some drawbacks: they can be computationally expensive and can produce audible side effects, among other things. Resampling, which is achieved by modifying the sampling rate, changes not only the pitch of the sound, but also its length and formant shape. The length change is usually compensated by a compression or expansion of the sound (Lee 1972), but this often introduces pops, clicks, and other modulations. The shift in the formant structure, which creates the familiar chipmunk-like sound, also causes the sound to be unrecognizable after only one octave or so of pitch shift. The analysis/resynthesis methods involving linear predictive filters (Oppenheim and Schafer 1975) and Fourier transforms (Dolson 1986) have the ability to retain the formant shape but require a large amount of computation. The goal of this paper is to present an alternative algorithm for altering the pitch and length of pitched (pseudo-periodic) sampled sounds. It has a computational efficiency near that of the resampling method while maintaining the formant characteristics of the unshifted sound.