Justin A. MacDonald, Phuong K. Tran
This paper details the results of an effort to develop a computer-based algorithm for sound localization for use in unmanned vehicles. The algorithm takes input from two or more microphones and estimates the position of the sound source relative to the microphone array. A priori knowledge of the stimulus is not required. The algorithm takes advantage of time-of-arrival and frequency cues to estimate the location of the sound source. The performance of two- and four-microphone implementations of the algorithm was measured using recordings from microphones mounted at several points around the head of an acoustic mannequin. Sounds were played at 5 degree intervals around the mannequin and the outputs were recorded. These recordings were fed into the algorithm that estimated the location of the sound source. Both algorithm implementations were able to identify accurately the location of a variety of real-world, broadband sounds, committing less than 2 degrees of unsigned localization error in moderately noisy environments. The four-microphone implementation of the algorithm was shown to be more resistant to background noise, committing less than 3 degrees of unsigned error when the signal-to-noise ratio was -10 dB or better. Future directions for algorithm development as well as potential applications are discussed.
Subjects: 1.6 Engineering And Science; 1. Applications