École Nationale Supérieure de
Techniques Avancées

Abstract

This report studies the problem of computer-assisted orchestration. A previous work proposed to find the exact result by a deterministic solution, and to compare it to the result found by "Orchidea", a software employing a deterministic algorithm, but giving the best acoustic results in the field. The general idea is to go through different sounds in a database, and find the subset of sounds which, when superimposed into a single sound, best reproduce the input sound, the so-called "target sound". The database is far too large to simply make all the possible combinations of sounds and compare them to the target sound. This report presents two possible solutions to the problem: a dynamic programming solution, and a linear problem formulation. For simplification purposes, the problem has been divided into four sub-problems of increasing difficulty. Each sub-problem is detailed for each of the two solutions. In a second step, this report presents how these problems were implemented and the difficulties they raised. It concludes with a presentation of the results of solving the problems in Python.