The timbre of violins is identified using machine learning, and a computer program is developed for the neural network using Python and Keras libraries. The 21 violins recorded include old Italian violins made by Stradivari and contemporary violins. The training and test data use the spectrum envelope and Mel-frequency cepstrum coefficients (MFCC). The accuracy of the identification test in the case of open strings is greater than 90%. Furthermore, experiments that predict similarity in timbre of an unknown violin to that of trained violins are presented.
Skip Nav Destination
December 22 2022
Identification of violin timbre by neural network using acoustic features
Masao Yokoyama, Yuya Ishigaki; Identification of violin timbre by neural network using acoustic features. Proc. Mtgs. Acoust. 11 September 2022; 49 (1): 035004. https://doi.org/10.1121/2.0001659
Download citation file: