Speaker
Description
Modeling of musical instruments is most often performed in a forward manner, arriving at an analysis or a sound from a chosen geometry and material parameters. But most often, the inverse problem is of more interest: how to build a musical instrument with a desired sound. Using machine-learning methods trained by sounds of physical model outputs when varying a geometrical or material parameter space, the overall timbre space of an instrument within reasonable variations is found. Varying the material parameters of a Dulcken cembalo of 1765 from the Museum of Applied Arts in Hamburg, the sound parameter space of this type of instrument is investigated. It is assumed that other Dulcken instruments from this builder have been built with the same geometry, but varying wood parameters, and therefore have sounded within such a timbral space. Comparing these variations of other cembalos in the future could lead to a characterization of different instruments within material parameter variations. This could lead to a reasonable reconstruction of historical instrument sounds.