Constrained optimization for binaural sound reproduction using a virtual artificial head

* Presenting author
Day / Time: 21.03.2018, 17:40-18:00
Room: Interim 2
Typ: Regulärer Vortrag
Session: Binauraltechnik
Abstract: Rasumow et al. (IEEE TASLP 2016) showed that spatial directivity patterns of individual HRTFs can be accurately synthesized in the horizontal plane using a microphone array with 24 microphones. To calculate the filter coefficients, a narrow band least-squares cost function incorporating a white noise gain (WNG) constraint was minimized. Fallahi et al. (DAGA 2017) improved the performance by adding constraints imposing boundaries on the synthesis error. However, the feasibility of finding a solution for this constrained-optimization problem depends on a variety of parameters in particular the array topology as well as the constraints. In this contribution, we modify the constraints until a feasible solution can be found, by relaxing the constraints on the spectral distortion at contralateral directions far from the median plane, by lowering the constraint on the WNG or by decreasing the spatial resolution at certain directions. Simulation results show that with only two parameters (one with respect to spectral distortion and one with respect to the WNG constraint) feasible solutions can be obtained up to high frequencies (8 kHz) in the horizontal plane. The next challenge is to address the constraints for different array topologies and for an increased number of directions in a 3-D scenario.