Calibration of a Microphone Array Based on a Probabilistic Model of Microphone Positions

Abstract

This paper addresses a novel method for calibrating microphone positions included in a microphone array. The performance of microphone array processing deteriorates due to two factors: (1) differences between predetermined position and actual positions of the microphones. (2) sound source signal overlaps in frequency and time. To solve these problems, we propose a probabilistic generative model of the sound propagation process determined by microphone and sound source positions. The model is defined as the product of three probabilities: (1) prior probability of the microphone positions based on reference positions, (2) prior probability of the sound source spectrum, and (3) conditional probability of the recorded spectrum. Based on the model, an iterative algorithm to calibrate the microphone positions is derived as a solution of the maximum a posteriori estimation. Preliminary experiments through numerical simulation with an 8-ch microphone array revealed that the proposed method accurately estimated the microphone positions when using multiple sound sources. Preliminary experiments through numerical simulation with an 8-ch microphone array suggested the proposed method accurately estimated the microphone positions when using multiple sound sources.