Analysis-synthesis strategies for simple and robust transformations of complex sounds

This project will explore the control of sound transformations applied to complex sounds such as: explosions with debris, glass breaking, various object impacts and bounces, etc. These complex sounds will be decomposed into multiple individual collisions components. Based on a database of simple collisions gathered from the analysis of one or several complex impact sounds, we will derive new techniques to control the transformation of model parameters in order to synthesize a multitude of sounds showing some characteristics of the originals with modified duration, density of collision, spectral and temporal domain evolution.

To achieve this, the project will build on well established signal segmentation techniques to perform a robust decomposition of complex sounds into its constituting elements. Improvements to traditional techniques for spectral envelope estimation will be applied to individual components identified and allow further transformation possibilities. Finally, common signal processing limitation to parametric modeling transformations will be addressed in order to provide transformation controls that can alter physical attributes of the collisions (e.g. material of the object, the point of impact, etc.).

Faculty Supervisor:

Philippe Depalle

Student:

Jung Suk Lee

Partner:

Audiokinetic Inc.

Discipline:

Other

Sector:

Information and communications technologies

University:

McGill University