A new sinusoidal model for synthesis of musical instruments

Abstract

Many algorithms have been developed over the years to synthesize acoustic sounds and are now used commercially in acoustic synthesizers and digital keyboard products. The issue with these algorithms is the tradeoff among sound fidelity, algorithm complexity, and hardware/storage requirements-the last two of which are directly related to the cost of the system. In this reported work, we introduce a new sinusoidal model for digitally synthesizing musical instruments. The new algorithm has unusually high fidelity, minimal memory requirements, and high computational efficiency. The algorithm is based on the “analysis-by-synthesis overlap add” (ABS/OLA) sinusoidal model, which models musical sounds as a short-time weighted sum of constant frequencies, phases, and amplitudes. The new model we introduce incorporates a novel dynamic pitch and frequency control feature in synthesis that allows very high quality instrument sounds to be generated over a wide range of pitches from a very short sampled recording of the musical instrument. Sound modifications can be performed parametrically within the framework all using fast Fourier transforms (FFTs) for high efficiency. Examples of synthetically generated non-western musical instruments will be presented during the conference and contrasted with competing technologies to illustrate the advantages of the new method.