Sound source localization with a gradient array using a coherence test
Satish Mohan, Michael E. Lockwood, Douglas L. Jones, Quang T. Su, Ronald N. Miles · The Journal of the Acoustical Society of America · 2003
The localization of sound sources is difficult when the number of sources is greater than the number of sensors. An algorithm is described that can localize in such acoustic scenes by utilizing a statistical coherence test to identify desirable time–frequency bins where the MUSIC and minimum-variance spectral estimators are applied. Localization results are then obtained by integrating across the selected time–frequency bins. The algorithm was used to localize signals in azimuth and elevation from a co-located microphone array consisting of one omnidirectional microphone and three gradient microphones. With 2.5 seconds of five speech sources, the azimuth and elevation estimates produced by the algorithm, over a hundred realizations of additive white Gaussian noise, had biases under four degrees and standard deviations less than two degrees. The algorithm was also used to compare the localization performance, in azimuth, of an array of two co-located gradient microphones with an array of two 15-cm-spaced omnidirectional microphones in the free-field.