Sound source localization using a distributed microphone system in real environments

Accurate localization of sound sources is indispensable for high-quality sound capture. The difference of distance from a sound source to two microphones causes a time delay of arrival (TDOA). The cross-power spectrum phase analysis (CSP) method has been proposed to estimate the TDOA based on correlation in signals. The sound source can be localized theoretically as a crossing point of sound directions estimated using two pairs of microphones. However, in a real environment, microphones receive sounds that have originated through reverberation and reflection. This situation makes TDOA estimation difficult, and accurate sound localization is difficult when using only two pairs of microphones. The authors propose a new method of sound localization based on multiple microphones that are widely distributed and placed under the ceiling of a room. A number of pairs of microphones give a series of CSP correlation coefficients as a function of the time delay. The CSP coefficients for hypothetical sound sources are accumulated over all the microphone pairs. The number of microphones is 16. The sound source is localized as a hypothesis that maximizes the accumulated CSP coefficient. Experiment results obtained in a real room show that the proposed method improves the localization accuracy.