Research

My main research area is computer music, and more precisely musical sound modeling.

The fundamental part of this research consists in spectral modeling. I am particularly interested in accurate analysis methods based on the Fourier transform. I have proposed a new method based on the sound signal derivatives, which lead to the demonstration of the equivalence of phase-based methods. I have also investigated fast additive synthesis: software oscillators, use of a psychoacoustic model, non-linear methods, and polynomial generators. This was done in the context of sinusoidal modeling, with extensions to non-stationary, hierarchic, as well as hybrid (deterministic + stochastic) cases.

These researches were generalized to computational auditory scene analysis (machine listening), and sound source separation using, for example, the harmonic structure or the spatial localization. The latter ("3D Sound") with a perceptive approach was particularly investigated, with the proposal of a simplified HRTF (Head-Related Transfer Function) model useful for both localization and spatialization, taking in account the azimuth and distance (and soon elevation) of each sound source. This allows a retroaction loop: a spatializer that listens to the produced sound and adjusts its spatialization parameters based on the observed error.

But the results of source separation turned out to be insufficient for demanding musical applications. However it is possible to enhance de quality of the results by injecting some information. I proposed a new research topic: informed solving of inverse problems, where partial information on the solution allows to control the quality of the result. Computer music is a great application domain for this, because this additional information is available through collaborations with artists. Thus, we investigated informed sound source separation (unmixing), as well as the inversion of the other steps of the whole music production chain: "musical decomposition", or audio reverse engineering. We made active listening come true from music fixed on some support (such as CDs), i.e. the possibility for the listener to modify the music while it is played (interaction), as the conductor of an orchestra or a disc jockey do. I was the national coordinator of the DReaM project on this topic.

I also conduct researches on computer graphics, and more precisely the analysis of images of archaeological objects. I have studied an image registration method in the spectral domain (based on the Fourier-Mellin transform), and its extension to 3D, by going from the image to the object model by photogrammetric techniques (or "shape from shading"). The IBISA project is an application of these computer vision techniques to the identification of archaeological objects (and more precisely of ancient coins, as an expert numismatist myself).

I aim at bridging the gap between these two research areas (sound and image), as in the case of the use of image analysis techniques for sound localization (Fourier-Hough transform).

My research it thus multidisciplinary, with sound (mainly) and image processing aspects, mathematical representations and psycho-physical considerations. But my approach is first of all the one of a computer scientist, focusing on algorithms, data structures, and complexity. I am interested in the computer modeling of sounds and images, especially for computer audition and vision for purposes of interaction, with applications to music and archaeology, respectively.