Quasi-real-time estimation of a loudspeaker direction from sound pressure level ratio among four channels

Abstract

This paper describes a quasi-real-time estimation method of loudspeaker direction under Gaussian noise environment. Our estimation method is based on sound pressure level ratio among four channels when the loudspeaker is surrounded by four microphones. Performance of our method tends to be better when acoustic spatial radiation characteristics of sound source is skewed. Preliminary experiments show that an offline processing version of our method is more than 99% accurate in acoustically clean environments. In this paper, we discuss the noise robustness of our method through simulation experiments. Loudspeaker direction is estimated from 0.34s segments of Japanese sentences every 0.01s, that is quasi-real-time processing. Those segments are contaminated by Gaussian noise at 0dB, 10dB, 20dB. We demonstrate that the proposed method is robust to Gaussian noise. The results show that accuracies of proposed method (Prop2) are 96.9%, 98.2%, 98.4% at the band the center frequency of which is 2000Hz.