The purpose of this paper is to bring about a unification of synthetic techniques for synthesizing sounds with desired spectra with sample-based ones. The synthetic techniques have the advantage of total specifiability, whereas the sample-based ones can be more natural sounding, and are much easier to wield.
Synthetic methods date at least to the invention of subtractive synthesis. In recent years, formant-generating systems have become more popular, starting with VOSIM  and continuing through FM formant generation , the FOF , and the PAF . Sample-based ones [5,6,7] have usually used pitch information to make phase coherent cross-fades between segments of recorded speech and singing.
Eckel et al. extended the FOF generator to FOG, in which sampled waveforms may replace the FOF's sinusoids. This has been further generalized to include PSOLA . In this paper we take a similar approach, extending an earlier effort to synthesize periodic sounds by concatenating wave packets. The work reported here is distinguished by careful attention to phase effects during cross-fades to produce desirable spectra even when the sampled signal deviates from periodicity. First we treat the case of purely sinusoidal ``grains", and then we discuss how to prepare recorded samples so that they can be used in the same way.