Parameter fitting for piano sound synthesis by physical modeling - Archive ouverte HAL Access content directly
Journal Articles Journal of the Acoustical Society of America Year : 2005

Parameter fitting for piano sound synthesis by physical modeling

Abstract

A difficult issue in the synthesis of piano tones by physical models is to choose the values of the parameters governing the hammer–string model. In fact, these parameters are hard to estimate from static measurements, causing the synthesis sounds to be unrealistic. An original approach that estimates the parameters of a piano model, from the measurement of the string vibration, by minimizing a perceptual criterion is proposed. The minimization process that was used is a combination of a gradient method and a simulated annealing algorithm, in order to avoid convergence problems in case of multiple local minima. The criterion, based on the tristimulus concept, takes into account the spectral energy density in three bands, each allowing particular parameters to be estimated. The optimization process has been run on signals measured on an experimental setup. The parameters thus estimated provided a better sound quality than the one obtained using a global energetic criterion. Both the sound’s attack and its brightness were better preserved. This quality gain was obtained for parameter values very close to the initial ones, showing that only slight deviations are necessary to make synthetic sounds closer to the real ones.
Fichier principal
Vignette du fichier
Bensa2005.pdf (221.89 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-00105865 , version 1 (04-09-2020)

Licence

Attribution

Identifiers

Cite

Julien Bensa, Olivier Gipouloux, Richard Kronland-Martinet. Parameter fitting for piano sound synthesis by physical modeling. Journal of the Acoustical Society of America, 2005, 118 (1), pp.495-504. ⟨10.1121/1.1929230⟩. ⟨hal-00105865⟩
79 View
88 Download

Altmetric

Share

Gmail Facebook Twitter LinkedIn More