著者
安藤 繁 篠田 裕之 小川 勝也 光山 訓
出版者
公益社団法人 計測自動制御学会
雑誌
計測自動制御学会論文集 (ISSN:04534654)
巻号頁・発行日
vol.29, no.5, pp.520-528, 1993-05-31 (Released:2009-03-27)
参考文献数
13
被引用文献数
5 21

This paper describes a theory, algorithm, and experimental results of a new 3-D sound source localization system which is based on the spatio-temporal derivative method applied to acoustic signals.The principle is outlined as follows. We observe the sound field through amplitude f and its spatiotemporal gradients fx, fy, ft of a single point. An instantaneous distribution of time-varying vectors fx, fy, ft, f in a 4-D gradient space is classified according to its degree of orderness. Each orderness is theoretically related to constraints on possible source locations. Under this classification, we obtain an unambiguous 3-D location of a single source under a second degree of orderness and a co-existing line or plane of double sources under a first degree of orderness.We applied the principle to a sound source visualizer system by which a possibility map of the source locale is accumulated and superposed realtimely on TV images. By several experiments, this system is shown to be able to localize multiple wideband sources such as white noises and human voices while discriminating a number of them.