Signal Independent Wideband Activity Detection Features for Microphone Arrays

Detection of activity is a key capability for microphone arrays. An array system should tell when a source of interest is present and evaluate the usability of the computed spatial estimates. This work proposes activity features that are computed from spatial data only, using time delays and direction of arrival. The features are validated with a loudspeaker experiment. Results show that the features are effective: tolerable detection errors are achievable with simple detection methods. In addition, direction of arrival estimation error is reduced down to one third when unreliable estimates are discarded