raoul hausmann, mechanical head : the spirit of our time
models and methods for describing spatial features of sound scenes – application to binaural recordings
doctoral thesis from the paris 6 university
doctoral school of mechanical sciences acoustics and electronics
this research aims at providing methods for automatically describing objective spatial features of a recorded sound scene, without prior knowledge of the sources, of the sound they emit, or the characteristics of the performance space. the study focuses specifically on estimating the direction of arrival of the sound event (assuming a unique and stable source) and on characterizing the power envelope of the reverberation. a theoretical framework, inspired from spatial hearing models, allows the development of a homogeneous set of methods for detection and estimation. these are based on nonstationary second-order statistics of relations between tracks. cues obtained in each frequency band are grouped together and interpreted by high-level descriptors. the relevance of this approach is studied on synthetic and real binaural recordings.