The analysis and processing of directional room impulse responses (DRIR) is a central theme in the team. The work focuses on the modeling and resynthesis of the late reverberation sound field. In this context, a robust denoising process has been developed to exploit the team’s DRIR database. When the late reverberation tail exhibits diffuse field properties (plane wave incoherence, isotropic spatial power distribution), denoising via resynthesis can be carried out in a spherical harmonics domain, the spatial representation at the root of the HOA format. The domain of spherical harmonics preserves the incoherent spatial nature of the diffuse field (thanks to the orthogonality of the decomposition base). The exponential energy decay is then modeled component by component and frequency by frequency, during which the noise threshold is simultaneously detected. This frequency-dependent noise floor may be then replaced by a synthesized incoherent tail parameterized by the DRIR energy decay envelope.
In the framework of a collaboration with Aalto University a study is devoted to the analysis of the directivity of the late field in the case where it is not entirely diffuse, i.e. where the plane waves are incoherent but the spatial power distribution is anisotropic. In this case, the spatial symmetry of the spherical harmonics cannot preserve anymore the spatial distribution of the reverberation tail. An alternative approach has been proposed based on plane-wave decomposition of the SRIR.
The number of the plane waves is limited by the spherical harmonics order to allow for an exact reconstruction. The spatial distribution of the plane waves is chosen by maximizing the angular difference between their steering directions (so that their signals are as independent as possible) and minimizing the overall directivity variance over the sphere. The denoising process is then applied in each direction which preserves the spatial incoherence and directivity of the reverberation tail. The following steps of this project are devoted to the extraction of high level DRIR descriptors in order to allow for the development of a collection of spatial transformations potentially interesting for artistic applications exploiting DRIRs.
IRCAM's Team : Acoustic and Cognitive Spaces
