Sound source separation in diesel engines can be implemented using a Wiener filter, or spectrofilter, that can extract the combustion contribution in the overall noise. This filter characterizes the transfer function between a cylinder pressure and a measurement point (vibratory or acoustic). An engine is characterized by several filters (one by cylinder) that are estimated for many operating conditions (engine speed and load). A purpose of this work is to obtain an average spectrofilter allowing the synthesis of the combustion noise in all operating conditions. This synthesis should be accurate enough to be used in perceptive studies. This approach requires an adapted processing, involving a short windowing of signals among other characteristics. Cyclostationarity properties beforehand enable to separate determinist and random components, in order to minimize bias error. However an important variability remains in the filters obtained for various operating conditions. This variability can be explained in the low frequency range because the random contributions are masked by measurement noise, which makes it difficult to separate combustion from mechanical noise. In the high frequency range, the acoustical response of the combustion chamber (that is part of the estimated filter) is strongly dependent on temperature because of its modal behaviour. In the intermediate part of the spectrum, the observed variability is relatively low, and can be partly imputed to the estimation errors. One of the objectives in this work consists in taking advantage of the multitude of information given by the estimations from different operating conditions in order to refine the spectrofilter estimation in this medium frequency band. To do this, an approach of experimental model is adopted, requiring the extraction of modal parameters from the great number of obtained filters. The impulse responses are decomposed on a complex exponential basis, using different approaches as ESPRIT method or LSCE (modal analysis). The spectrofilters estimated from different operating conditions are analyzed and compared in this reduced basis, in order to identify the underlying structural parameters. These parameters are compared to the results of an experimental characterization of the non running engine. The level of detail in the synthesis (number of components of the filter) is another important factor that has to be studied, but it depends on the final application. This work is realized to aiming to implement perceptive studies of the combustion noise. Another potential application concerns the structural health monitoring of the engine, that can be supervised by examining the spectrofilter evolution.