DETERMINING A SCENE GEOMETRY FROM ITS IMPULSE RESPONSE David García-Garzón; Daniel Arteaga; John Usher; Toni Mateos Barcelona Media – Centre d’Innovació | Av. Diagonal, 177, planta 9, 08018 Barcelona david.garcia@barcelonamedia.org ; daniel.arteaga@barcelonamedia.org; john.usher@barcelonamedia.org; toni.mateos@barcelonamedia.org Abstract This paper reports on the development of a system to characterize the geometric properties of a space from an acoustic impulse response measured within it. This can be thought of as the inverse problem to the common practice of obtaining impulse responses from either a real-world or a virtual space. Starting from a measured or synthesized impulse response in an original scene, the method described here defines a space of possible scenes and a distance measure on it, and discusses a strategy to select a scene from this space, which is perceptually as close as possible to the original one. Potential applications of this novel method include audio forensics, re-equalization and re-mixing of music and audio. Keywords: Room impulse response, room geometry, source localization. 1 Introduction This paper concerns the development of a system to characterize the physical properties of a space from an acoustic impulse response measured within that space, be it real or virtual. This is in contrast to the common practice of generating an impulse response from either a real-world space, or from a virtual space. In other words: rather than generating an acoustic impulse response from a given room, we generate a room from a given impulse response (IR); and hence this task is called the “Inverse IR” problem. 1 Indeed, the research presented here attempts at inferring not only the room geometry and the property of the materials present in it, but also the position of the sound sources and receivers. We will therefore refer to such as Scene Inference From Audio (SIFA) algorithms, whereby the concept of scene refers to the more generic setup described above. 1 The “inverse IR problem” was named in an unpublished report by B. Katz and his workgroup: http://www.limsi.fr/Rapports/RS2005/chm/ps/ps12/ 1