Large-Scale Physical Modeling Synthesis, Parallel Computing, and Musical Experimentation: The NESS Project in Practice

Abstract

Sound synthesis using physical modeling, emulating systems of a complexity approaching and even exceeding that of real-world acoustic musical instruments, is becoming possible, thanks to recent theoretical developments in musical acoustics and algorithm design. Severe practical difficulties remain, both at the level of the raw computational resources required, and at the level of user control. An approach to the first difficulty is through the use of large-scale parallelization, and results for a variety of physical modeling systems are presented here. Any progress with regard to the second difficulty requires, necessarily, the experience and advice of professional musicians. A basic interface to a parallelized large-scale physical modeling synthesis system is presented here, accompanied by first-hand descriptions of the working methods of five composers, each of whom generated complete multichannel pieces using the system.

Physical Modeling, Algorithms, and Sound Synthesis: The NESS Project

“…This leads to great simplifications, especially when dealing with connections between objects, and also yields computational structures suitable for parallelization. For more on parallelization aspects of the sound synthesis methods presented here, see the companion article (Bilbao et al 2020).…”

“…Because execution times are relatively small, brass synthesis has been a mainstay for composers using the NESS system (cf. Bilbao et al 2020).…”

“…For a given input, then, there will be a natural degree of coherence among the various outputs, and thus a holistic approach to spatialization is possible (see Figure 10). For more on the use of such environments in a multichannel setting, see the companion article in this issue of Computer Music Journal (Bilbao et al 2020).…”

“…Work on the first two, at the level of models and algorithm design, and across a wide variety of instrument types, is described in this article. Work on the third difficulty, through implementation strategies in parallel hardware and, to a much more tentative level, the fourth, is detailed in a companion article in this Journal (Bilbao et al 2020). This article is intended for a relatively nontechnical audience.…”

Automatic Synthesis Technology of Music Teaching Melodies Based on Recurrent Neural Network

“…AI has already been adopted to realize algorithm composition or automatic music generation [8][9][10][11][12]. Bilbao et al [13] introduced bidirectional long short-term memory (LSTM) neural network to the mixed music generation system and thus realized the training of multi-voice music datasets. eir approach provides effective chord progressions while ensuring melody time and transposition invariance.…”