Music analysis/synthesis by optimized multiple wavetable interpolation

• Authors: Xiaobo Li;Jonathan Jeffrey James Mohr
• Venue: Music analysis/synthesis by optimized multiple wavetable interpolation
• Year: 2002

Multiple wavetable interpolation is a form of music analysis/synthesis that involves three basic steps: (1) The recorded sound is reduced to a set of break-points by piecewise linear approximation of the spectral envelopes of its harmonics; (2) the spectrum at each breakpoint is matched by determining weightings for a small number of wavetables; and (3) the sound is resynthesized using multiple wavetable additive synthesis by interpolating between the weightings for each wavetable at consecutive breakpoints. This thesis presents a new analysis/synthesis method, optimized multiple wavetable interpolation, that generalizes and optimizes multiple wavetable interpolation. The method uses a clustering algorithm to select a bank of wavetables such that the wavetables will be useful in matching the breakpoint spectra of a wide variety of harmonic tones played by various instruments. The breakpoint-matching algorithm selects subsets of the wavetables in the wavetable bank that best match each breakpoint spectrum, subject to the constraint that a wavetable that ceases to be used at a given breakpoint must be faded out by the next breakpoint and one that comes into use must be faded in. This algorithm introduces the use of the single-source acyclic weighted shortest path algorithm to choose breakpoint matches in a globally optimal way. The output of the algorithm is a sequence of n-tuples of pairs of wavetable indices and weights which can serve as a control stream for a hardware or software synthesizer. A secondary contribution of this research is a new breakpoint-selection algorithm which operates by segment merging; the algorithm has characteristics that make it well suited to use on instrumental tones, especially those with vibrato or other pronounced amplitude changes.