Blind 2D signal direction detection with limited sensor space using maximum likelihood estimation

Humans can automatically turn toward the origin of a sound that they hear through their both ears. This capability is crucial in daily life. A computer that can recognize the direction of the source of a sound would also be useful for many types of applications. This paper describes the development of blind two-dimensional (2D) signal direction recognition using two recording sensors with limited space between the sensors in a noise-free environment. The proposed 2D source direction recognition method is based on time-delay estimation using maximum likelihood estimation by forming a histogram of a power-weighted spectrum corresponding to attenuation and time-delay index. A histogram-boundary method is also proposed that relates to a distance of the two microphones. In addition, the optimal number of time-index bins for the histogram were investigated. Given the narrowness of the space between microphones (i.e., 3.2 centimeters), the proposed method can acceptably estimate the position of the sound source. In experimental testing on real audio sources, the proposed method has demonstrated a higher level of directional performance than an existing method.