A real-time implementation for sound source localization inside a spherical microphone array
* Presenting author
In the ongoing research project NarDasS, we developed a system for binaural reproduction of self-generated sound in virtual acoustic environments (VAEs). A 32-channel surrounding spherical microphone array captures the direction-dependent sound of an acting user. Based on these captured signals and specific BRIRs (BRIR – Binaural Room Impulse Response), a room response is then calculated by dynamic binaural synthesis in real time, and auralized via headphones. The system is designed for a sound source located in the center of the array. However, in real life, the user and in particular the sound source might be slightly off-center. To compensate for this offset, the microphone signals need to be adjusted in level according to the 1/r distance law. This requires information on the exact position of the sound source inside the array. This work presents a C++ implementation for sound source localization in real time. The time differences of arrival (TDOAs) between the microphones are calculated by a cross-correlation with phase-transform weighting, and a linear equation system is set up. This equation system is then solved by the least-square method. In a technical evaluation, we analyze the localization accuracy for several test cases and real-life scenarios.