Block-sparse approach for the identification of complex sound sources in a room

Geometrical acoustic softwares are necessary to produce auralizations for specific sound environment. Whether the room impulse response computation require point-source and receiver to be of omnidirectional sensitivity, the influence of their directivity on the resulting virtual audio rendering Is relevant. It is then crucial to account for when simulating accurately a calibrate acoustic model.We treats here the case of the source directivity. The use of a spherical surrounding microphone array remains the most natural way to measure it. The source is located inside the delimiting volume. The radiated pressure is sampled at fixed points. The directivity pattern is then computed in term of spherical harmonics functions. But due to hardware complexity, most of the spherical antennas in the litterature have a few number of microphones. This limits the performance of the antenna in term of resolution and bandwidth. Also, decomposition errors can appear with a possible mismatch between the acoustic center of the source and the origin of the array. An additional optimization task is required which increases the complexity of the process.In this paper, we propose a practical strategy, comprising a dedicated algorithm and an array design, to estimate the directivity pattern of complex sound sources. The study takes place in reverberant rooms.Firstly, we describe a greedy-sparse algorithm called Block Orthogonal Matching Pursuit. By this iterative approach, the identification and characterization tasks can be joint in a unique scheme. This facilitates the acoustic center research. However, under non-anechoic conditions, BlockOMP fails because of the free-field propagation assumption. Considering the first reflections to approximate the room transfer function permits to solve the inverse problem. The notion of virtual microphone arrays, based on an analogy with the Image Source Method, is introduced to extend the validity of BlockOMP. Numerical results supply a proof of the concept in a scenario including multiple acoustic sources.Secondly, a large three-dimensional microphone array is deployed. Largeness concerns here in both its dimensions and the number of microphones. The array consists of five sub-planes which surround the entire room where sources are located. The acquisition system comprises digital MEMS microphones chips. The entire signal processing chain is directly integrated on the captor. The microphones are flush mounted on the walls of the room. The true location of the sensors is known, given by an acoustic geometrical calibration step. The 1024 MEMS record synchronously the pressure signal emitted by the sources. From each harmonic spectral component, a sparse spherical harmonics decomposition of each target can be achieved.An experiment is performed to assert the efficiency of the proposed strategy. The goal is to recover the nature of two prototypes of source. They are build from an unbaffled loudspeaker, arranged to show a dipole and a quadripole behaviour. Their directivity pattern are previously measured under controlled conditions using a semi-circular array of 64 microphones. This database serves here as reference. For the experiment, they emit the same signal simultaneous. The results with our system indicates good correlations. Separating both the sound radiating contribution is well achieved. Our last study case deals with the voice directivity measurements. If the dependence with the frequency has been established, the effect of the phonema variation is rarely identified. We demonstrate here that our apparatus constitutes a powerful tool to examine this aspect.