Representing the Timbre of Traditional Musical Instruments Based On Contemporary Instrumental Samples Using DDSP

Abstract

his project explores the potential of Differentiable Digital Signal Processing (DDSP) to represent and synthesize the timbre of five different notes of the Korean traditional musical instrument, Geo- mungo, using digital instrumental samples of the bass guitar, which has a similar mechanism to produce the sound. To evaluate the fea- sibility and quality of the digital recreation process, we compared hand-played Geomungo audio samples with digitally recreated au- dio samples using DDSP. The MFCC, spectral contrast, chroma features, and raw signal comparison, were used for assessment. Our findings show the possibility of applying DDSP to represent and synthesize the nuances of pitch and dynamics for expressive aspects of Geomungo’s five different notes effectively. We also pro- pose three audio features that can be used to evaluate the results quantitatively under the context of neural sound synthesis.