Proceedings of the 18 th Sound and Music Computing Conference, June 29 th – July 1 st 2021 100 REAL-TIME I MPLEMENTATION OF THE S HAMISEN USING F INITE DIFFERENCE S CHEMES Titas LASICKAS(tlasic16@student.aau.dk) 1 , Silvin WILLEMSEN(sil@create.aau.dk) 2 , and Stefania SERAFIN(sts@create.aau.dk) 2 1 CREATE, Aalborg University, Copenhagen, Denmark 2 Multisensory Experience Lab, CREATE, Aalborg University, Copenhagen, Denmark ABSTRACT The shamisen is a Japanese three-stringed lute. It is a chor- dophone that has the front of the body covered by a ten- sioned membrane which greatly contributes to the distinct sound of the instrument. Although the shamisen is a tra- ditional Japanese instrument, it is a rare instrument in the rest of the world, making it mostly inaccessible by the ma- jority of artists. To our knowledge, no physically modelled synthesizer of the shamisen is available, forcing produc- ers and musicians to use samples. The objective of this paper is to make the shamisen’s distinct sound more acces- sible to digital music artists. The real-time implementation of the shamisen physical model is presented along with the derivation of solution using the finite-difference time- domain (FDTD) methods. The digital instrument sounds mostly as intended, though lacking the shamisen’s distinct buzzing sound requiring further development. 1. INTRODUCTION The shamisen (see Figure 1) is a Japanese three-stringed lute, with origins in China. This instrument is a chordo- phone that has a membrane covering its soundbox; this stretched membrane contributes to the distinct sound of the instrument. The instrument is played using a bachi, a large plectrum which is held in one hand, while using the fingers of the other hand to pinch the strings against the neck of the instrument, allowing the user to play different pitches. The timbre of the shamisen has a very distinct buzzing which is associated with a low nut which lets a vibrating string come in the contact with the neck [1] and the specific playing technique that increases the percussiveness of the instru- ment by hitting the body with the bachi during the plucking of the strings [2]. The main goal of this paper is the digitalization and real- time simulation of an instrument that, due to its rarity, is not easily available to the majority of artists. Currently, the only method of obtaining shamisen sounds without having the instrument at hand is by using a sample of the instru- ment [4] or one of the audio plugins [5–7] which are also based on a sampled shamisen. Usually, a single sample Copyright: © 2021 the Authors. This is an open-access article distributed un- der the terms of the Creative Commons Attribution 4.0 International License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Figure 1. The shamisen (taken from [3]). of a note played at a single pitch is pitch-shifted enabling artist to play melodies and chords. Such a method relies on the recordings being done in an anechoic chamber to reduce the effects of the room response. Moreover, due to only usually recorded one note sample, the pitch shifting can introduce artifacts. Even when the sampling is done for all possible pitches on each string, the user is stuck with the way the performer played the instrument during the recording. To rectify these undesirable qualities of the sampling method and to create more scope for skillful ar- ticulation when performing, the shamisen can be simulated using a physical model instead. Although several stringed musical instruments have been mostly simulated using digital waveguides [8–10], in this paper we model the shamisen by using Finite Difference Time Domain (FDTD) methods [11]. FDTD methods re- quire developing a full mathematical description of the sys- tem. Such a description development uses partial differ- ential equations which are discretized using FDTD meth- ods, yielding finite difference schemes (FDSs). FDTD methods provide better spatial accuracy when the model has frequency-dependent damping and dispersion [12, 13]; in addition, FDTD methods are more flexible as no as- sumptions are being made about the linearity of the so- lution [11]. Alternatively, a modal approach – such as in [14, 15] – could be used as it is generally much more efficient. Additionally, modal synthesis for 2D systems appeared in [16,17], however, to retain generality for con- trol and easier implementation when connecting multiple models, FDTD methods are chosen here . In addition, the nonlinear collision of the bachi with the membrane can be modelled straight-forwardly using FDTD methods. The real-time implementation of similar models, modelled using FDTD approach have been achieved by the authors of [18], where a real-time banjo is recreated using a field programmable gate array and by the authors of [19] where