Predicting Timbre Features of Instrument Sound Combinations: Application to Automatic Orchestration

Abstract

In this paper we first introduce a set of functions to predict the timbre features of an instrument sound combination, given the features of the individual components in the mixture. We then compare, for different classes of sound combinations, the estimated values of the timbre features to real measurements and show the accuracy of our predictors. In the second part of the paper, we present original musical applications of feature prediction in the field of computer-aided orchestration. These examples all come from real-life compositional situations, and were all produced with Orchide´e, an innovative framework for computer-aided orchestration recently designed and developed at IRCAM, Paris.

Computer-Aided Musical Orchestration Using an Artificial Immune System

“…The sounds s (i) in an orchestration S (i) should approximate the target s T when played together. Therefore, the combination functions estimate the values of the spectral features of S (i) from the features of the isolated sounds s (i) normalized by the RMS energy e (i) [6]. The combination functions for the spectral centroid μ (i), spectral spread σ (i), and loudness λ (i) are given respectively by…”

“…Compositional experimentation in orchestration arises from the increasing tendency to specify instrument combinations to achieve desired effects, resulting in the contemporary use of timbral combinations [15,18]. The development of computational tools that aid the composer in exploring the virtually infinite possibilities resulting from the combinations of musical instruments gave rise to computer-aided musical orchestration (CAMO) [3][4][5][6]11,17,18]. Most of these tools rely on searching for combinations of musical instrument sounds from pre-recorded datasets to approximate a given target sound.…”

“…To overcome the drawbacks of subtractive spectral matching, Carpentier and collaborators [3][4][5][6]19] search for a combination of musical instrument sounds whose timbral features best match those of the target sound. This approach requires a model of timbre perception to describe the timbre of isolated sounds, a method to estimate the timbral result of an instrument combination, and a measure of timbre similarity to compare the combinations and the target.…”

“…In metric MDS spaces, the distance measure directly allows timbral comparison. Models of timbral combination [6,12] estimate these features for combinations of musical instrument sounds.…”

“…Carpentier and collaborators [3][4][5][6]19] consider the search for combinations of musical instrument sounds as a constrained combinatorial optimization problem [5]. They formulate CAMO as a variation of the knapsack problem where the aim is to find a combination of musical instruments that maximizes the timbral similarity with the target constrained by the capacity of the orchestra (i.e., the database).…”