Reprinted from Proceedings, ICMC 2005, pp. 733-736.
Many techniques have been proposed for synthesizing the sung or spoken voice; they generally fall into two classes. First, the purely synthetic techniques (FM, FOF, PAF, ...) make sounds with specified formant frequencies, bandwidths, and amplitudes. Second, the sample-based methods (LPC, Vocoders, PSOLA, ...) reconstruct sounds from pre-recorded speech or singing. Here we present a technique that can synthesize spectra either with specified formants, or in imitation of a recorded sound. Since it is not patent encumbered it can be used to replace the PAF operator for open-source re-implementations of pieces using PAFs.