Predicting timbral and perceptual characteristics of orchestral instrument combinations

Abstract

Orchestration is a compositional practice that consists of writing for several instruments. This process often involves harnessing each instrument's sound to create sonic textures that could not be achieved with a single instrument. These sound fusions are usually sought by composers to express specific perceptual effects. However, the number of potential combinations is significant. Testing and analyzing all combinations to identify the ones matching the desired perceptual effects is logistically and computationally complex. Using supervised learning methods to create regression and classification models, it is possible to predict specific timbral and perceptual characteristics from information about a combination of different orchestral instruments. Such developments would provide methods to estimate the perception of instrument timbre fusions directly from abstract information. Similar methods could potentially be applied to other types of sources and predict specific perceptual characteristics without the need to perform an acoustical and psychoacoustical analysis on every audio sources.Orchestration is a compositional practice that consists of writing for several instruments. This process often involves harnessing each instrument's sound to create sonic textures that could not be achieved with a single instrument. These sound fusions are usually sought by composers to express specific perceptual effects. However, the number of potential combinations is significant. Testing and analyzing all combinations to identify the ones matching the desired perceptual effects is logistically and computationally complex. Using supervised learning methods to create regression and classification models, it is possible to predict specific timbral and perceptual characteristics from information about a combination of different orchestral instruments. Such developments would provide methods to estimate the perception of instrument timbre fusions directly from abstract information. Similar methods could potentially be applied to other types of sources and predict specific perceptual characteristics without...