ScalableWavetable Matching for Real-Time Polyphonic Synthesis

Recent work on wavetable synthesis of musical instrument tones assumes that a constant number of wavetables are used throughout the duration of a tone. A scalable approach would dynamically allocate wavetables to simultaneous voices to allow more polyphony and improve the sound quality. Iterative wavetable matching finds the basis spectra one by one over several iterations, and offers real-time control of the number of wavetables, supporting scalability. This paper describes and evaluates several iterative methods. Matching results for a range of instruments show that, on average, iterative local search can find matches with errors within 0.5% of nearoptimal non-iterative solutions. Iterative local search only rarely gets stuck on bad local optima.