An evaluation of hardware-software design for sound source localization based on SoC

Abstract

The article presents an evaluation of sound source localization algorithm based on field-programmable gate array (FPGA) and ARM processor. The proposed architecture performs the computation of azimuth and elevation angles of the speaker location in a real-time. Time-difference of arrival (TDOA) features are computed for a grid with four microphones. The speaker position is estimated by matching the features with the acoustic maps. The influence of line and square placement of the microphones is evaluated. The average azimuth errors for square grids with aperture 0.4 m, 0.6 m, 0.8 m are up to ± 7◦, ±5◦ and ±4◦ respectively. Line grid with 0.4 m step and square grids with 0.6 m and 0.8 m aperture give average ̄ǫel = ±4◦ in a range 45◦ < αreel < 90◦ at 2 m distance from grid to speaker.