”Time-frequency analysis and granular synthesis ofengine noise”

摘要

The noise in the car compartment is tightly related with one of the major car noise sources - the car engine. For the automotive constructors, the optimization and thesound design requires a simulation and prediction of the car noise, often used also as appli-cation in psychoacoustics for parameters evaluation. We propose a signal model for enginenoise simulation, based on the granular synthesis concept. In this hybrid model are takeninto account some main physical phenomena in the engine's cycle and, on the other hand,their corresponding sounds from a signal point of view. The continuous linear time-frequencyanalysis with an appropriate gaussian window is first used to detect the sound events in theengine's cycle. Each event is then decomposed on an optimized number of elementary signals(atoms), so the linear combination of these atoms gives the reconstructed signal. Several criteria for estimation and optimization of the synthesis parameters are established, to keep the most important information on the initial signal in a minimum number of coefficients, in order to achieve a perceptually satisfying noise synthesis. The granular synthesis model gives good preliminary results for the simulation of the engine noise at low speed. These results encourage to extend the use of this granular model to some more generic driving situations.