Interface[s] Montréal

Digital Sound

Philippe Depalle

McGill University, Faculty of Music

Spectral Models of Audio Signals. Application for the Synthesis and Processing of Sounds

View the presentation [In French]

Computerised sound synthesis is a multidisciplinary research area that relates to the study of ways to generate sound from symbolic (most often), mathematical and/or computerised representations. The interest in the use of synthesis for the creation of music resides essentially in its potential to generate new or unheard-of sounds that may not be produced by actual devices (instruments or other). Whatever the musical end result, the scientific process used in its synthesis includes: the formulation of representation and production models for sound signals (modeling), the extraction of synthesis parameters for recorded signals (analysis of sound signals), the development of tools (essentially in terms of software implementation), and the generation of control parameters for tools (synthesis control). For several years now, synthesis models have been classified into two categories, often referred to as signal models and physical models. Physical models relate to the representation of sound generating devices, i.e. the cause or the instrument.

During this presentation, we will limit our scope to the description of signal models that represent the perceptive properties of the sound signal without reference to the characteristics of the instrument that produced it. Signal models describe the sound effect, as opposed to physical models, which describe the cause. In practice, signal models are comprised of an abstract structure (signal processing type), designed to code the information linked to the perceptive effects within the structure parameters. More precisely, most signal models are based on a representation of spectral and code information, presented in a relatively condensed format, relating to the frequency, phase and amplitude characteristics of sound signals (Example: additive synthesis, phase vocoder, source-filter type synthesis, etc.).

Following an introduction on the concept of modeling in the sound synthesis context, we will concentrate on the analysis-synthesis methodology based on signal models. For musicians, this methodology unifies the concepts of sound synthesis and processing; while for scientists, research is focused primarily on the analysis portion. The primary challenge in studying sound signals is having the ability to accurately track their spectral progression. The presentation will be illustrated through sonic/acoustic examples as well as descriptions of practical applications.

Philippe Depalle has been Associate Professor of Music Technologies at the Schulich School of Music at McGill University since 1999. He previously held the positions of Associate Professor at École Superieure d’Électricité, Metz, France (1985-1988); Researcher and member of the Analysis/Synthesis team at IRCAM, Paris, France (1988-1997); and Visiting Professor for the Faculty of Music at Université de Montréal (1997-1999). His research activities and interests relate to the analysis, processing and synthesis of sounds, the simulation of musical instruments, and the control of spectral synthesis and analysis. His contributions in these fields are varied and include: publications; contributions to the development of musical software applications (AudioSculpt, Additive and Diphones); and a patent on additive synthesis techniques. The fundamental component of his body of work is the systemization of the analysis/synthesis approach used in the design of computer music tools. One mass-audience application of this approach was the reconstitution of a Castrato voice for music in the film – Farinelli, released in 1994.